Methods and systems for determining values of ventilatory parameters

Abstract

The present disclosure relates to methods and systems for determining exercise related ventilatory data of a person (8), in particular determining one or more ventilatory thresholds of the person (8) and / or determining a ventilatory zone of the person (8). Microphone signals received from a microphone (2A, 2B, 2C) associated with the person (8) are received and analyzed to determine the breathing rate, and using the breathing rate the ventilatory threshold(s) and ventilatory zone(s) are determined.

Description

[0001] METHODS AND SYSTEMS FOR DETERMINING VALUES OF VENTILATORY PARAMETERS

[0002] FIELD OF THE DISCLOSURE

[0003] The present disclosure relates to methods and systems for determining values of ventilatory parameters. In particular, the present disclosure relates to determining values of ventilatory parameters including a ventilatory threshold and / or a ventilatory zone of a person using a breathing rate as determined from an acoustic signal measured by a microphone.

[0004] BACKGROUND OF THE DISCLOSURE

[0005] Changes in the physiological state of the human body during periods of exertion, for example physical exertion, are expressed and can be determined by measuring physiological parameters. These physiological parameters are indicative, for example, of cardiovascular changes and metabolic changes and include the heart rate and parameters related to heart rate variability, respiratory rate and volume, the body skin and core temperature, blood lactate concentration, and oxygen consumption and carbon dioxide production, as well as further indicators derived from a combination of one or more of the aforementioned.

[0006] As the intensity and / or duration of exertion increases, the physiological state continuously changes, and many models, using for example thresholds, scales and / or zones, have been proposed with which a current physiological state can be readily gauged and / or classified. While models may rely solely on a perceived exertion of the person, more sophisticated models use measured physiological parameters to determine which particular physiological state the person currently inhabits.

[0007] Examples of models include the perceived exertion scale (using a scale of 1 - 10 or 6 - 20, for example). Other models employ a measured heart rate which is then used, typically together with a maximum heart rate and optionally a minimum heart rate, to determine a heart rate zone. Common models may use from three zones up to seven zones or more, with a first zone typically being considered a “recovery” zone during which exertion is minimal, an “endurance zone”, all the way up to an “anaerobic” or “maximal exertion” zone.

[0008] Measuring and recording the physiological parameters and using these to determine a performance zone has become increasingly important to plan and optimize training and for improving performance during competition, and for tracking performance in general.

[0009] Tracking heart rate, for example using photoplethysmography (PPG) sensors or electrocardiography (ECG) sensors worn on the wrist or across the chest, for example, is commonly used for recreational athletes. However, measuring solely the heart rate for measuring and classifying exertion has a number of draw-backs. Firstly, the heart rate tends to lag behind the actual exertion by up to 30 seconds, and therefore does not provide immediate feedback. Secondly, the heart rate is just one of multiple vital signs which change depending on physical exertion... Therefore, ambitious and professional athletes additionally use other physiological signals for determining their current performance level. This may include blood lactate measurements, which is an important indicator of exertion and performance, which however typically require invasive and frequent blood sampling and analysis. Another method uses cardiopulmonary exercise testing (CPET), as discussed below.

[0010] Another group of physiological parameters are related to respiration, as these provide additional information related to the physiological state of the person during exertion and are also linked to the accumulation of lactate in the blood and a respiratory state. These include respiration rate, ventilation volume, oxygen consumption (often normalized by body weight), and carbon dioxide production. These physiological parameters however can currently only be measured directly in a laboratory setting as the person is required to wear a face mask with tubes connected to equipment for measuring the stated physiological parameters.

[0011] The respiration rate can be used to measure and classify the level of exertion. According to some methods, two ventilatory thresholds, VT 1 and VT2, are used to classify the level of exertion into one of three zones, the first zone Z1 being below VT 1 , the second zone Z2 being between VT 1 and VT2, and the third zone Z3 being above VT2. If enough time is spent at a high enough intensity in the third zone, VO2 max may be reached, i.e. the person will reach a state of maximal oxygen uptake. It is also possible to use more than 3 zones, by splitting according to suitable criteria, such as e.g. certain percentage ranges of the second ventilatory threshold.

[0012] As exercise intensity increases from rest, the breathing rate generally remains relatively steady up until the first ventilatory threshold VT 1 . However, it should be noted that the slope of the breathing rate varies individually and a slow increase can sometimes be observed already below VT1. Up until VT 1 , the blood lactate concentration also remains more or less at its baseline value and breathing remains easy and unlabored - the person is able to talk more or less comfortably and complete a sentence without feeling the need to breathe. The increased demand in oxygen due to the exercise is primarily met by an increase in the tidal volume, i.e. the volume of air breathed per minute, and not the respiration rate.

[0013] The first ventilatory threshold VT 1 represents a level of exercise intensity at which the blood lactate begins to accumulate above its baseline. In the process to buffer acid metabolites, which accumulate together with the lactate, the body produces additional CO2 that is exhaled through an increase in ventilation volume and often also an increase in breathing rate. Breathing becomes more labored and it is increasingly difficult to speak comfortably (i.e. complete a sentence without taking a breath). Between VT1 and VT2, the blood lactate production increases in a roughly linear manner with exercise intensity. VT 1 is also indicative of a level of exercise intensity at which the majority of fuel is from fats (also called fat-max). Training at, or just below, VT1 has been shown to be beneficial for improving aerobic performance in athletes as it is a level of exercise intensity which can be maintained for prolonged durations, and frequently, without generating undue fatigue. Between VT 1 and VT2, the breathing rate also tends to increase with increasing exercise intensity, however the degree of increase varies between individuals.

[0014] VT2 marks the second point beyond which the respiration rate increases disproportionately to the exercise intensity and is sometimes called the respiratory compensation threshold (RCT). Talking becomes very difficult. At this point, the blood lactate concentration begins to accumulate faster than the body is capable of clearing it.

[0015] Techniques for indirectly measuring some of the physiological parameters related to respiration are known, for example the respiratory rate can be measured indirectly by measuring a heart rate signal, as small variations in the heart rate are caused by the expansion and contraction of the lungs, a phenomenon called respiratory sinus arrhythmia. Since these variations occur in the range of several tens of milliseconds and are very faint, they cannot be determined reliably using PPG sensors in a wrist band or sports watch while moving due to motion artefacts, and the use of an ECG monitor worn across the chest is required. Further, breathing rate may be measured by bio-impedance, however this has the disadvantage that electrodes must be placed on the body.

[0016] SUMMARY OF THE DISCLOSURE

[0017] It is an object of the disclosure and embodiments disclosed herein to provide methods, devices and systems for determining values of ventilatory parameters, in particular exercise related ventilatory data of a person, more particularly determining one or more ventilatory thresholds of the person and / or determining a ventilatory zone of the person. In particular, it is an object of the disclosure and embodiments disclosed herein to provide a computer-implemented method, and an electronic system including one or more electronic devices for determining values of ventilatory parameters, in particular exercise related ventilatory data of a person, more particularly determining a breathing rate, determining one or more ventilatory thresholds of the person and / or determining a ventilatory zone of the person, which does not have at least some disadvantages of the prior art.

[0018] At least some of the steps described with reference to one of the objects of this disclosure, in particular a first method, may also be performed as part of another of the objects of this disclosure, in particular a second method, and vice versa.

[0019] The present disclosure relates to a method for determining a breathing rate of a person. The method comprises receiving, in a processor, a recorded microphone signal from a microphone associated with a person performing exercise, wherein the microphone signal was recorded during a recording period including at least two periods of different exercise intensity. The method comprises determining, in the processor, breathing rate values of the person from the microphone signal. The breathing rate values of the person may be used to determine a ventilatory threshold, a ventilatory zone, and / or other physiological parameters, values, or indicators of the person.

[0020] The present disclosure relates to a method for determining a ventilatory threshold of a person. The method comprises receiving, in a processor, a recorded microphone signal from a microphone associated with a person performing exercise, wherein the microphone signal was recorded during a recording period including at least two periods of different exercise intensity. The method comprises determining, in the processor, breathing rate values of the person from the microphone signal. The method comprises analyzing, in the processor, the breathing rate values of the person, in particular two or more breathing rate values during the at least two periods of different exercise intensity, respectively, to determine a ventilatory threshold indicative of a separation between two ventilatory zones of the person. The ventilatory threshold may be defined as a particular value of a breathing rate.

[0021] Depending on the embodiment, one or a plurality (in particular, two) ventilatory thresholds may be determined. Thereby, two or more ventilatory zones may be identified.

[0022] The method provides for a simple procedure for determining one or more ventilatory thresholds which does not require directly measuring blood lactate levels or respiratory gas (e.g., rate of oxygen uptake or carbon dioxide production), such as in CPET. The ventilatory threshold(s) thus established may be used in any number of useful and productive ways, particularly for monitoring, managing or improving exercise performance during training or competition by ambitious athletes, but also by recreational athletes or people performing exercise for improving their cardiovascular abilities and health. The ventilatory threshold(s), as indicated by breathing rate(s), are also by themselves useful indicators of cardiovascular performance and ability.

[0023] The breathing rate values are a plurality of indicators of a breathing rate, the breathing rate defined, for example, as a number of breaths per minute. The breathing rate values may also be a smoothed representation of the directly sampled breathing rate, for example by using a moving average.

[0024] One of the periods of exercise intensity may be a rest period. The rest period, at which the person is at rest or performing very light exercise, may be used to determine a resting breathing rate. One of the periods of exercise intensity may be a period of moderate exercise intensity, and one of the periods of exercise intensity may be a period of high intensity. The exercise intensity may be defined objectively, for example based on a resistance level defined by exercise equipment, or based on a performance level as determined by sensors. The exercise intensity may, additionally or alternatively, be defined subjectively, for example using a rating of perceived exertion (RPE), for example based on a scale of 1 - 10 or 6 - 20, or based on a descriptive scale using terms such as easy, moderate, hard, very hard, and maximum effort.

[0025] In an embodiment, the method comprises receiving, in the processor, an exercise type which identifies a type of exercise performed by the person. The method may comprise processing the microphone signal according to the exercise type. In particular, the method may comprise determining the breathing rate values of the person depending on the exercise type. Additionally or alternatively, the method may comprise determining the ventilatory threshold depending on the exercise type. The exercise type may be received in the processor based on input by the person (e.g., via a human-machine interface) or may be automatically detected by the processor using sensor signals.

[0026] Thereby, the method enables for processing of the microphone signal, determination of the breathing rate, and / or determination of the ventilatory threshold(s) specific to the type of exercise. This is beneficial because different exercise types place different demands on the body which may result in different breathing rate values with respect to exercise intensity and / or the ventilatory threshold. Additionally, different exercise types can result in different noise profiles in the microphone signal which require signal preprocessing specific to the exercise type.

[0027] The exercise type is, for example, running (e.g., outdoor running on track, tarmac and / or trail, or indoor running on a treadmill). The exercise type is, alternatively or additionally, cycling (e.g., outdoor cycling on a bicycle or indoor cycling on a cycling ergometer), rowing (e.g., outdoor rowing on a rowing boat, or indoor rowing on a rowing ergometer), and / or skiing (e.g., outdoor skiing or indoor skiing on a skiing ergometer, downhill skiing, cross-country skiing). Depending on the embodiment, the method may further comprise generating, in the processor, a message indicative of the ventilatory threshold(s), respectively. The message may be stored in a memory. The message may be transmitted to another device. The message may also be provided to the person by way of a human-machine-interface.

[0028] Depending on the embodiment, the human-machine-interface may provide the message using visual, auditory, or haptic signals, as described herein in more detail.

[0029] In an embodiment, the one or more ventilatory thresholds are determined, in the processor, using the breathing rate values and a ventilatory threshold model trained using machine learning. The ventilatory threshold model may include, for example, an AdaBoost optimized decision tree. The ventilatory threshold model may be trained using supervised learning, for example by use of a ventilatory threshold training dataset including breathing rate values from a large number of individuals, the ventilatory threshold training dataset including breathing rate values and manually, automatically or semi-automatically labeled ventilatory thresholds. In an embodiment, alternatively or additionally to the microphone signal, a recorded or live-streamed acoustic transducer signal may be received from an acoustic transducer. The acoustic transducer is a device configured to convert vibrations received through a medium into an electrical signal. The medium may include air, skin, or bone. For example, the acoustic transducer may be implemented as an ear bone microphone, bone-conduction microphone, and / or digital stethoscope.

[0030] The microphone may be carried by the person, worn by the person, or more generally in possession of the person. The microphone may be, for example, a lapel microphone, a hand-held microphone, a headset microphone, an audio cable with a microphone, a headphone microphone, a training device microphone embedded, integrated or connected with a sports computer or sports equipment (stationary or moveable) used by the person, a hearing aid microphone, an earbud microphone, etc. It may also consist of an array of multiple microphones.

[0031] The microphone may also more generally be arranged in proximity to the person, in particular in a proximity such that the microphone is able to record breathing related sounds of the person.

[0032] Depending on the embodiment, a recorded signal from a plurality of microphones may be received. Thereby, a combined microphone signal may be used for determining the breathing rate more accurately, the combined microphone signal increasing a signal to noise ratio, for example by the method using spatial filtering techniques for reducing background noise.

[0033] More generally, the recorded signal may be received from one or more transducers. The transducers include a microphone but may further include an accelerometer, for example.

[0034] In an embodiment, the ventilatory thresholds may be determined by using a difference in the breathing rate during exercise between a first period at a first exercise intensity and a second period at a second exercise intensity. For example, the first period at the first exercise intensity may be a rest period, for example having a duration of between 30 seconds and 5 minutes, such that a resting breathing rate may be determined. The second period at the second intensity may be a period of moderate exercise intensity at which the breathing rate becomes elevated above the resting breathing rate.

[0035] The pre-defined exercise protocol may also include a final period of rest or recovery at no or a low level of intensity, in order to determine a recovery breathing rate, respectively. In an embodiment, a first ventilatory threshold may be determined as an increase of the breathing rate above a resting, or baseline, breathing rate by a defined margin. A second ventilatory threshold may be determined as an increase of the breathing rate above the baseline breathing rate and / or above the first ventilatory threshold by further defined margin(s).

[0036] In an embodiment, the method further comprises receiving, in the processor, an exercise protocol indicator associated with a pre-defined exercise protocol, the pre-defined exercise protocol defining two or more levels of exercise intensity at which the person is to exercise for the two or more periods. The method comprises determining, in the processor, the one or more ventilatory thresholds using the exercise protocol indicator. The pre-defined exercise protocol may be, for example, a ramp test protocol in which the exercise intensity increases from an initial intensity step-wise or continuously until the person reaches volitional exhaustion. The pre-defined exercise protocol may alternatively be a set of intervals of one or more defined intensities with rest periods in between. The pre-defined exercise protocol may simply define an exercise intensity at which the person exercises until volitional exhaustion. The exercise protocol indicator may be input by the person.

[0037] In an embodiment, the method further comprises detecting, in the processor, the predefined exercise protocol, for example from amongst a plurality of pre-defined exercise protocols. The processor may detect the pre-defined exercise protocol using the breathing rate values and / or using additional information, in particular one or more performance signal(s), one or more movement value(s) and / or one or more physiological values received from one or more external devices.

[0038] In one embodiment, the pre-defined exercise protocol is a ramp protocol with a steady, i.e. nearly continuous, increase in the resistance level. In another embodiment, the predefined exercise protocol is a ramp protocol with a plurality of discrete levels of resistance, and the duration of each step of the ramp is between 30 seconds and 2 minutes, wherein durations of longer than 1 minute are preferred such that the breathing rate may stabilize for a particular resistance level. During the ramp protocol, the breathing rate values will tend to steadily increase.

[0039] In an embodiment, the method comprises detecting the pre-defined exercise protocol as a ramp protocol if the performance signal(s), movement signal(s) and / or physiological value(s) received during at least part of the exercise is indicative of a steady increase in exercise intensity.

[0040] In an embodiment, the one or more ventilatory thresholds are determined by fitting the breathing rate values, or smoothed breathing rate values, using a piecewise function, in particular a piecewise linear function with two or three pieces. The changepoints, i.e. the points at which the slope changes, are indicative of the one or more ventilatory thresholds. The method includes selecting the changepoints as those points which result in an optimal residual (e.g., a smallest residual sum of squares or least squares).

[0041] Additionally or alternatively, the changepoints are selected using a regression decision tree, in particular a model tree.

[0042] Additionally or alternatively, the changepoints may be selected or determined using a regression model. For example, a random forest may be used as part or whole of the regression model.

[0043] The breathing rate may be directly associated with the level of resistance at the same time-point, or the breathing rate may be associated with an earlier level of resistance (this is because the breathing rate may lag behind a changing level of resistance by, for example, 30 seconds - 2 minutes, depending on the level of intensity and the cardiovascular abilities of the person). In an embodiment, the pre-defined exercise protocol is configured such that the person traverses (crosses) at least one ventilatory threshold, preferably two. Thereby, the person is in at least two of the ventilatory zones during exercise.

[0044] In an embodiment, the method further comprises: generating, in the processor, a control signal for an auxiliary exercise device, the control signal defining one or more levels of resistance generated by the auxiliary exercise device according to the pre-defined exercise protocol, the control signal thereby defining the two or more levels of exercise intensity. The method comprises transmitting, by the processor, using a communication interface connected to the processor, the control signal to the auxiliary exercise device. The method comprises determining, in the processor, for each of the one or more ventilatory thresholds, an association with at least one of: particular level of resistance or a particular level of exercise intensity at which the ventilatory threshold is reached.

[0045] The auxiliary exercise device is, for example, a treadmill, stationary bike, rowing ergometer, or similar exercise equipment in which the resistance provided to the person can be controlled via a control signal. The level of resistance provided can be defined according to the type of auxiliary exercise device. For example, on a treadmill, the level of resistance may be defined using a speed and / or an inclination of the treadmill. On a stationary bike, the resistance may be defined as a power which the person is required to provide, and on the rowing ergometer, the level of resistance may be defined as a particular rowing pace.

[0046] Depending on the embodiment, the auxiliary exercise device comprises a manual control to adjust the resistance, rather than, or in addition to, control via the control signal.

[0047] By determining an association between a ventilatory threshold and a level of resistance, the person and / or the auxiliary exercise equipment may define subsequent exercise intensity relative to the ventilatory threshold to optimize training. For some types of exercise, for example, training just below, at, or just above a ventilatory threshold may elicit beneficial physiological training adaptations.

[0048] Depending on the embodiment, the method may further comprise generating, in the processor, a message indicative of an association between the ventilatory threshold(s) and particular level(s) of resistance, respectively. The message may be stored in a memory. The message may be transmitted to the auxiliary device. The message may also be provided to the person by way of a human-machine-interface.

[0049] In an embodiment, the method further comprises receiving, in the processor, during the exercise, a performance signal from an external device associated with the person, the further signal indicative of a measured performance level of the person during exercise. The method comprises associating, in the processor, using the further signal, the measured performance level with the one or more ventilatory thresholds. The method comprises generating, in the processor, a message indicative of one or more performance levels associated with the one or more ventilatory thresholds, respectively.

[0050] The performance signal may include, for example, a heart rate. In this case, the external device may include a heart rate monitor. The performance signal may include a power. In this case, the external device may include an ergometer (e.g., a cycling ergometer, rowing ergometer). The performance signal may be a speed, pace or resistance level. The external device in this case may include a GPS receiver.

[0051] By determining one or more associations between the one or more measured performance levels and corresponding one or more ventilatory thresholds, the person may optimize subsequent exercise, for example by exercising at a specific heart rate associated with a particular ventilatory threshold, or exercising at a particular power or pace associated with the particular ventilatory threshold. Depending on the embodiment, the method may further comprise generating, in the processor, a message indicative of an association between the ventilatory threshold(s) and particular performance level(s), respectively. The message may be stored in a memory. The message may be transmitted to the external device. The message may also be provided to the person by way of a human-machine-interface.

[0052] In an embodiment, the method further comprises determining, in the processor, ventilation volume values of the person from the microphone signal and analyzing, in the processor, the breathing rate values and the ventilation volume values to determine the one or more ventilatory thresholds of the person. The ventilatory thresholds may therefore be defined using a defined function of the breathing rate and the ventilation volume (or a proxy of the volume). For example, the ventilatory threshold may be defined such that it is considered reached once the breathing rate has reached or exceeded a particular value of the breathing rate and the ventilation volume has also reached or exceeded a particular value of the ventilation volume. Naturally, more complex relationships are also possible.

[0053] The ventilation volume may be defined using an absolute value, for example a volume of air inhaled and / or exhaled. The ventilation volume may also be defined as a relative value, for example relative to a baseline value established when the person is at rest, in particular when the breathing rate is at a resting breathing rate value.

[0054] In an embodiment, the ventilation volume is determined using a signal amplitude from a transducer. The signal amplitude may be directly measured or inferred. As described herein, the transducer may be a microphone, however the use of other transducers together with, or as an alternative to the microphone is also foreseen. In particular, the transducer may be designed to measure a chest expansion of the person and provide a chest expansion signal. In an embodiment, the ventilation volume is determined, in the processor, using a signal amplitude of the microphone signal, in particular the signal amplitude during a breath. The signal amplitude, indicative of the audio volume during breathing, correlates with the ventilation volume. Further, the ventilation volume may be determined using a change in timing of the breaths, the timing indicative of breaths being long and deep or shallow and short, for example. The timing of the breaths may further relate to a rhythm of breathing in and breathing out. Further, the ventilation volume may be determined using a power distribution of the microphone signal, in particular during a breath. For example, a power distribution with a relatively greater proportion of higher frequency is correlated with an increase in ventilation volume.

[0055] In an embodiment, the method further comprises receiving, from one or more external devices, one or more environmental and / or physiological values. The method comprises using the received one or more environmental and / or physiological values to determine the ventilatory threshold(s). The environmental values may include, for example, an ambient temperature, an ambient humidity, and / or a barometric pressure. The physiological values may include, for example, a skin temperature, a core temperature, a sweat rate and / or a sweat sodium concentration.

[0056] In addition to the method for determining the one or more ventilatory thresholds of a person, the present disclosure also relates to a method for determining a ventilatory zone of a person from among a plurality of ventilatory zones. The ventilatory zones are separated by one or more pre-defined ventilatory thresholds. The method comprises receiving, in a processor, a microphone signal from a microphone associated with a person. The method comprises determining, in the processor, a breathing rate of the person from the microphone signal. The method comprises determining, in the processor, a ventilatory zone of the person, from among the plurality of ventilatory zones, using the breathing rate and the one or more pre-defined ventilatory thresholds. In an embodiment, the method comprises generating, in the processor, a message indicative of the determined ventilatory zone. The message may be stored in a memory. The message may be transmitted to the auxiliary device. The message may also be provided to the person by way of a human-machine-interface.

[0057] The pre-defined ventilatory thresholds may be determined using the method for determining the one or more ventilatory thresholds described herein. The pre-defined ventilatory thresholds may alternatively, or prior to performing the method for determining the ventilatory threshold(s), be defined using default values. For example, the first ventilatory threshold may have a value of between 15 and 25 breaths per minute, and the second ventilatory threshold may have a pre-defined value of between 20 and 40 breaths per minute. The pre-defined ventilatory thresholds may alternatively, or prior to performing the method for determining the ventilatory threshold(s), be defined using a defined relation between personal information of the person and ventilatory thresholds, the personal information including, for example, sex and / or age. Additional personal information may include weight and / or height. Further personal information may include a subjective and / or objective assessment of cardiovascular fitness, frequency at which the person performs exercise, and indications of disease or other medical issues.

[0058] In an embodiment, the method further comprises determining, in the processor, ventilation volume values of the person from the microphone signal and analyzing, in the processor, the breathing rate values and the ventilation volume values to determine the ventilatory zone of the person. Each ventilatory threshold may be pre-defined as a function of both the breathing rate and the ventilation volume. For example, if a particular ventilatory threshold is defined as having been reached or exceeded if both the breathing rate is above a defined value for the breathing rate and the ventilation volume is above a defined value for the ventilation volume, then the processor may determine the person to be in a ventilatory zone above that particular ventilatory threshold if both the breathing rate and the ventilation volume are both above their defined values for that particular ventilatory threshold, respectively.

[0059] In an embodiment, the method further comprises pre-processing, in the processor, the microphone signal using a digital filter, wherein the digital filter is configured to isolate, in the microphone signal, breathing-related sounds. Or in other words, the digital filter is configured to improve the signal to noise ratio by removing particular frequencies or frequency bands, in a time-dependent and / or dynamic fashion, such as to reduce background noise. Thereby, the breathing rate and optionally the ventilation volume can be determined more accurately. The digital filter may include two or more filters in parallel, such that the microphone signal is filtered differently for determining the breathing rate than for determining the ventilation volume.

[0060] In an embodiment, the digital filter may comprise a band pass filter, wherein the band pass filter isolates signals from approximately between 0.3 Hz to 1.25 Hz. Thereby, the breathing rate, which is typically between 20 to 75 breaths per minute during exercise, is isolated and high frequency noise reduced.

[0061] In an embodiment, the method may further comprise receiving movement data from a movement sensor. The method may include pre-processing the microphone signal further using the movement data. The pre-processing may be implemented as part of the digital filter.

[0062] The movement sensor may be include an inertial motion unit, for example including one or more accelerometers and / or a gyroscope. The movement sensor may include a GPS receiver. Thereby, for example, the microphone signal may be pre-processed, to remove motion related acoustic artefacts, such as the sound of steps and / or moving arms. In an embodiment, the method may further comprise determining, in the processor, the breathing rate using the movement data. For example, the method may include analyzing the movement data to determine a movement frequency (e.g., a step frequency, pedaling frequency, or rowing frequency) and / or a movement timing (e.g. time-points of when then foot strikes the ground). The method may then comprise determining a correspondence between the movement frequency and / or a movement timing. The method may then, if a correspondence is determined, determine the breathing rate using the movement frequency and / or the movement timing. For example, the breathing rate may be interpolated and / or extrapolated, for example in situations where the breathing rate may not be determined accurately using the microphone signal. This may be particularly beneficial for some persons whose breathing rate (i.e. breathing rhythm)is entrained with their movement rate (e.g., their movement rhythm, such as running rhythm, cadence, or step frequency). Entrainment is said to occur when there is an integer ratio between the breathing rate and the movement rate, for example in that they breathe once for a given number of steps, e.g. one breath per three steps, or breath three times for every five steps.

[0063] In an embodiment, the method comprises receiving, in the processor, a position of the microphone on the person. The method comprises isolating, in the microphone signal, breathing-related sounds using the position of the microphone.

[0064] Depending on the embodiment, the processor may determine the position of the microphone, for example by detecting the type of microphone or wearable / peripheral device into which the microphone is integrated or connected to. Additionally or alternatively, the processor may determine the position of the microphone or wearable / peripheral device by receiving, from the person, an input indicative of a position of the microphone or wearable / peripheral device. In an embodiment, the method comprises receiving, a plurality of microphone signals from a plurality of microphones, and pre-processing the plurality of microphone signals to isolate a facial region of the person, for example using the received position of the microphone. The pre-processing may be implemented as part of the digital filter, in particular as part of a spatial filter included in the digital filter.

[0065] In an embodiment, determining the breathing rate from the microphone signal comprises identifying, in the processor, using a breathing analysis model, a breath by analyzing a defined time-window in the microphone signal. The method comprises determining the breathing rate using a sequence of defined time-windows, preferably an overlapping sequence of defined time-windows, across a defined period of the microphone signal, and identifying unique breaths in the defined period.

[0066] The breathing rate may then be determined, for example by counting the unique breaths in the defined time period or using the mean / median of the average adjacent breath distances in the defined period to infer the rate.

[0067] In an embodiment, the method may comprise determining, in the processor, time-points of inhalations and / or time-points of exhalations. The method may comprise using the time-point of an inhalation and / or the time-point of the exhalation to infer the breathing rate. In particular, a median time between exhalations may be used to infer breathing rate.

[0068] In an embodiment, the method comprises determining the ventilation volume using a duration of an inhalation and / or a duration of an exhalation. The method comprise determining the ventilation volume further using an acoustic volume of an inhalation and / or an acoustic volume of an exhalation. In particular, the acoustic volumes of the inhalation and / or exhalation may be defined relative to a baseline acoustic volume, for example established during rest. Breaths may be identified further assuming a defined regularity in breathing, in particular that the breathing rate does not change faster than a particular rate of change. Interpolation to assume a breath has taken place if noise is too high or signal is missing etc.

[0069] In an embodiment, the method comprises analyzing, in the processor, the defined timewindow using a breathing analysis model. The breathing analysis model may be trained using supervised learning and a labeled training dataset. The training dataset includes a plurality of recorded microphone signals from a plurality of people and / or a plurality of trials. The labels may identify the presence and / or the absence of a sound event in an audio snippet (i.e., a short sample of a recorded microphone signal). The labels may additionally or alternatively identify a time-point of an onset and / or a time-point of an offset of a sound event in the audio snippet.

[0070] In an embodiment, the breathing analysis model is configured and / or trained to determine the ventilation volume.

[0071] For example, the breathing analysis model may include a regression network configured to determine peak-valley distance.

[0072] In an embodiment, the breathing analysis model is configured to use a pre-defined acoustic fingerprint of breathing to determine breathing rate values from the microphone signal. The pre-defined acoustic fingerprint of breathing is configured such that individual breaths in the microphone signal may be identified. For example, the pre-defined acoustic fingerprint may define a representative time-series of a signal indicative of a breath, and / or a representative spectrogram of a breath. The pre-defined acoustic fingerprint may, alternatively or additionally to the aforementioned, comprise a summary, simplification and / or representation of the representative time-series and / or spectrogram. For example, the pre-defined acoustic fingerprint may comprise an envelope function and / or parts thereof indicative of a breath. For example, the predefined acoustic fingerprint may comprise one or more representative frequencies and / or frequency bands which may further be defined by relative amplitudes.

[0073] For example, the breathing analysis model may be configured to sweep across the microphone signal, which may be a recording of the microphone signal or a real-time stream of the microphone signal and identify, and, using for example techniques such as pattern recognition, identify occurrences in the microphone signal indicative of a breath. The breathing analysis model may be configured to determine particular breathing time-points within the microphone signal. The breathing analysis model may be configured to determine the duration of a breath. The breathing analysis model may be configured to determine a rate of breaths and thereby determine breathing rate values.

[0074] Depending on the embodiment, the breathing analysis model may be configured to identify a breath by identifying parts of a breath, for example an inhalation and / or an exhalation. For example, the pre-defined acoustic fingerprint of breathing may include a pre-defined acoustic fingerprint of an inhalation and / or a pre-defined acoustic fingerprint of an exhalation. The breathing analysis model may be configured to identify nasal breaths and / or oral breaths. For example, the pre-defined acoustic fingerprint of breathing may include a pre-defined nasal breathing fingerprint and / or a pre-defined oral breathing fingerprint.

[0075] The breathing analysis model may include further acoustic fingerprints of other sounds, for example coughing or sneezing. The breathing analysis model may be configured to disregard a part of the microphone signal in which these other sounds are identified.

[0076] In an embodiment, the pre-defined acoustic fingerprint is generic, applicable to all persons. The pre-defined acoustic fingerprint may be determined using a training dataset including a plurality of sample breaths which are preferably identified and / or labeled. The pre-defined acoustic fingerprint may be determined by averaging across a plurality of the sample breaths.

[0077] In an embodiment, the pre-defined acoustic fingerprint can also be person-specific, i.e. applicable to the particular person. The pre-defined acoustic fingerprint may be generated during an initialization phase. The microphone signal received during the initialization phase, in which the person records their breathing in a preferably quiet environment, is analyzed to generate the pre-defined acoustic fingerprint. For example, the acoustic fingerprint may be generated using a rules-based approach. In a simple implementation, a repeating waveform which occurs at rates of approximately between 10 times per minute and 75 times per minute is assumed to be due to breathing. The acoustic fingerprint is then generated using a plurality of such waveforms, for example as an average. Alternatively or additionally, the acoustic fingerprint may be generated using a data driven approach or other suitable method to select appropriate features in the microphone signal which correlate with breaths or parts of breaths.

[0078] In an embodiment, the pre-defined acoustic fingerprint can also be group-specific, where the group may be based on age, age-range, gender, fitness level or another criterion.

[0079] In an embodiment, the method further comprises identifying, in the processor, using the breathing rate values and the one or more pre-defined ventilatory thresholds, a ventilatory transition indicative of the breathing rate crossing one of the one or more predefined ventilatory thresholds.

[0080] The ventilatory transition may be determined by comparing a current breathing rate of the person with one or more recent past values (e.g. a value 5 seconds past, 10 seconds past or a moving average of past values (i.e. 5-10 seconds in the past could be used) of the breathing rate, as well as the ventilatory thresholds. Thereby, a single outlier would not cause the ventilatory transition to be determined.

[0081] A ventilatory transition may be determined across a last ventilatory threshold that was crossed if the current breathing rate exceeds or falls below the last ventilatory threshold that was crossed and has a difference with respect to the last ventilatory threshold that was crossed by more than a defined margin (the defined margin may be expressed as a particular breathing rate, for example 2 - 5 breaths per minute). This provides for a measure of hysteresis, such that a person is not determined according to the method to hop “back and forth” across the last ventilatory threshold that was crossed if their breathing rate is fluctuating about the last ventilatory threshold only by a small amount. The defined margin may depend on a time elapsed since the last ventilatory threshold was crossed, i.e. since the last ventilatory transition was determined. For example, the defined margin may be relatively larger if the time elapsed since the last ventilatory threshold is relatively smaller. For example, within 30 seconds of a ventilatory transition being determined, the defined margin may be 5 breaths per minute, however, if the time elapsed since the last ventilatory threshold was crossed is between 30 seconds and 1 minute, the margin may be 2 breaths per minute.

[0082] In an embodiment, the method further comprises generating a message indicative of a ventilatory transition having occurred. The message may be stored in a memory. The message may be transmitted to another device. The message may also be provided to the person by way of a human-machine-interface.

[0083] In an embodiment, the method further comprises determining, in the processor, a physiological load on the person during exercise, the physiological load determined using the breathing rate values and a period or duration of the exercise. In particular, the physiological load may be determined as a function of the breathing rate values during exercise. The physiological load may be determined, for example, as being proportional to an average breathing rate multiplied by the duration of the exercise. Other methods may be used for determining the physiological load, in particular integrating the breathing rate values across the duration of exercise. The breathing rate values may be normalized, for example using a resting or baseline breathing rate. The breathing rate values may also be weighted, for example exponentially weighted such that greater values of the breathing rate contribute proportionally more to the physiological load than smaller values of the breathing rate.

[0084] In an embodiment, the method comprises determining the physiological load including an aerobic physiological load and / or an anaerobic physiological load. The aerobic physiological load may be calculated, by the processor, using one or more periods during exercise in which the person is above a baseline breathing rate and below a particular ventilatory threshold, in particular the second ventilatory threshold (the second ventilatory threshold being at a larger breathing rate than the first ventilatory threshold). The anaerobic physiological load may be calculated, in the processor, using one or more periods during exercise in which the person is at or above the second ventilatory threshold.

[0085] In an embodiment, the method further comprises determining, in the processor, a performance level and / or performance zone of the person using the ventilatory zone of the person and further physiological data. The further physiological data includes one or more of a heart rate, a power, a speed, a blood lactate value, an oxygen consumption and / or a carbon-dioxide production.

[0086] In addition to the method for determining one or more ventilatory thresholds and the method for determining a ventilatory zone of the person, the present disclosure also relates to a method for generating a training dataset designed to train a breathing analysis model. The method comprises receiving, in the processor, from each of a plurality of test persons (i.e. , test subjects), a recording of a microphone signal from a microphone and a recording of a dynamic chest expansion signal from a chest monitoring device configured to measure an expansion of the chest. The method comprises determining, in the processor, for each dynamic chest expansion signal recording, a plurality of identified breathing time-points using the dynamic chest expansion signal. These identified breathing time-points may be considered as ground-truth breathing time-points. The method comprises generating, for each microphone signal recording, a plurality of labels indicative of breathing time-points, using the identified breathing timepoints. Thereby, an automatically labeled training dataset is generated using which the breathing analysis model may be trained.

[0087] In an embodiment, the method comprises determining a periodicity of the dynamic chest expansion signal, in particular in a range of 0.3 Hz to 1.25 Hz, to determine identified breathing time-points. Additionally, the method comprises applying a constraint to the identified breathing time-points, the constraint restricting the rate of change of the breathing rate to a physiologically plausible rate of change of unforced breathing.

[0088] The sampling rate of the chest expansion signal may be in the range of 10 Hz - 50 Hz.

[0089] In an embodiment, the method comprises determining, for a particular identified breathing time-point (as identified using the dynamic chest expansion signal), a confidence level indicative of a measure of confidence for the particular identified breathing time-point. The measure of confidence may be in the form of a score, for example in the range of 0 to 1 , a label or a flag, such that the identified breathing timepoint may be manually checked and / or labeled as a breath.

[0090] In an embodiment, the method further comprises receiving, from the chest monitoring device, a bio impedance signal, and determining the plurality of identified breathing timepoints further using the bio impedance signal. The bio impedance signal measured at the chest also varies periodically according to the breathing, and therefore the bio impedance signal provides a further independent source of information for identifying the breathing time-points, thereby increasing the reliability of identifying breathing timepoints. For example, a breathing time-point may be identified if both the bio impedance signal and the dynamic chest expansion signal are indicative of a breath.

[0091] In an embodiment, the method further comprises receiving, from the chest monitoring device, movement data, in particular from an inertial measurement unit (IMU) integrated in or connected to the chest monitoring device. The movement data may be used to identify and / or label footsteps, in particular identifying time-points associated with footsteps. The time-points indicative of footsteps may be associated with the microphone signal. Further, artifacts in the dynamic chest expansion signal may be detected and removed. The IMU data may also be used to clean up the dynamic chest expansion signal (e.g. bending down to tie laces can cause a chest contraction to be registered even without a breath taking place)

[0092] In an embodiment, the method further comprises initializing a breathing analysis model and training the breathing analysis model using supervised training and using the (labeled) training dataset.

[0093] The breathing analysis model may include, for example, a random forest model and / or a neural network. The neural network may include one or more fully connected layers, convolutional layers, attention layers, or a combination thereof.

[0094] The present disclosure also relates to a training dataset suitable for training a breathing analysis model. The training dataset includes a plurality of recordings of microphone signals from a plurality of persons, respectively. The training dataset further for each person a recording of a dynamic chest expansion signal, which was recorded simultaneously to the microphone signal. The training dataset may further include, for each person, a simultaneous bio impedance signal representative of chest movement. In addition to the methods and the training dataset described herein, the present disclosure also relates to an electronic system comprising a processor. The processor is configured to perform at least one of the methods as described herein. The electronic system may be centrally arranged or distributed. Further, the processor may be part of an electronic device worn by or carried on a person, or the processor may be part of an electronic device arranged remotely.

[0095] In an embodiment, the electronic system further comprises a microphone worn on or held by the person, the microphone configured to generate the microphone signal.

[0096] In an embodiment, the electronic system includes a wearable device and the processor is integrated into the wearable device. The wearable device may be implemented as, for example, a smart watch or sports watch.

[0097] In an embodiment, the electronic system includes a device associated with the person configured to generate a performance signal indicative of a measured performance level of the person. The device may include a heart rate sensor. The device may further include an inertial measuring unit (IMU), for example including one or more accelerometers and / or a gyroscope. The device may further include a GPS receiver. The device may be implemented as a stationary exercise device, e.g., a rowing erg, cycling erg, treadmill, etc.

[0098] In an embodiment, the electronic system includes a server computer and the processor is integrated into the server computer.

[0099] The present disclosure also relates to a computer program product comprising computer program code configured to control a processor such that the processor performs at least one of the methods described herein. The present disclosure also relates to a non-transitory computer readable medium comprising computer program code configured to control a processor such that the processor performs at least one of the methods described herein.

[0100] The present disclosure also relates to a data product comprising a training dataset suitable for training a breathing analysis model to determine breaths in a microphone signal, wherein the breathing analysis model includes labeled breaths in a plurality of recorded microphone signals from a plurality of persons. The breaths may have been manually labeled, or labeled using physiological data indicative of a breath, for example dynamic chest expansion signals and / or bio-impedance signals of the plurality of persons. The data product may be stored on a non-transitory computer readable medium.

[0101] BRIEF DESCRIPTION OF THE DRAWINGS

[0102] The herein described disclosure will be more fully understood from the detailed description given herein below and the accompanying drawings, which should not be considered limiting to the invention described in the appended claims. The drawings in which:

[0103] Fig. 1 shows a diagram illustrating a person jogging wearing a headset including a microphone, as well as a wearable device;

[0104] Fig. 2 shows a block diagram of an electronic system;

[0105] Fig. 3 shows schematically a microphone communicatively connected to a smart phone;

[0106] Fig. 4 shows schematically a microphone communicatively connected to a wearable device; Fig. 5 shows schematically a microphone communicatively connected to a server computer via a smart phone;

[0107] Fig. 6 shows schematically a wearable device with a microphone communicatively connected to a server computer;

[0108] Fig. 7 shows a flow diagram illustrating a method for receiving and processing a microphone signal;

[0109] Fig. 8 shows a flow diagram illustrating a method for analyzing breathing rate values to determine one or more ventilatory thresholds;

[0110] Fig. 9 shows a flow diagram illustrating a method for associating a resistance level and / or intensity level with one or more ventilatory thresholds;

[0111] Fig. 10 shows a flow diagram illustrating a method for associating a performance level with one or more ventilatory thresholds;

[0112] Fig. 11 shows a flow diagram illustrating a method for determining a ventilatory zone using a breathing rate;

[0113] Fig. 12 shows a flow diagram illustrating a method for determining a breathing rate of a person;

[0114] Fig. 13 shows a flow diagram illustrating a method for generating a training dataset suitable for training a breathing analysis model;

[0115] Fig. 14 shows a flow diagram illustrating a method for training a breathing analysis model;

[0116] Fig. 15 shows a flow diagram illustrating a method for determining a performance zone using the breathing rate and heart rate; Fig. 16 shows two charts in which the breathing rate and the heart rate, respectively, are plotted as a function of time and power during a ramp test;

[0117] Fig. 17 shows two charts in which the relative signal amplitude and spectrogram of a microphone signal including breaths are shown over several seconds; and

[0118] Fig. 18 shows a depiction of a breathing analysis model implemented as a neural network.

[0119] DESCRIPTION OF THE EMBODIMENTS

[0120] Reference will now be made in detail to certain embodiments, examples of which are illustrated in the accompanying drawings, in which some, but not all features are shown. Indeed, embodiments disclosed herein may be embodied in many different forms and should not be construed as limited to the embodiments set forth herein; rather, these embodiments are provided so that this disclosure will satisfy applicable legal requirements. Whenever possible, like reference numbers will be used to refer to like components or parts.

[0121] Fig. 1 shows a human body of a person 8 performing exercise, in particular running. Illustrated are a number of possible positions 81 , 82, 83 where a microphone 2A, 2B, 2C may be arranged, in particular worn. Depending on the embodiment, one or more microphones 2A, 2B, 2C may be foreseen, the microphones arranged at one or more of the indicated positions 81, 82, 83. It is also foreseen that, at one particular position 81 , 82, 83, multiple microphones 2A, 2B, 2C may be arranged. The positions 81 , 82, 83 include a chest area, in particular a lapel position 81, an arm area, in particular a wrist position 82, and a head area, in particular an ear position 83. Other positions are possible, in particular an upper arm area, a face or more particularly mouth area, etc.

[0122] The microphone 2A, 2B, 2C is preferably worn close to the face such that breathing related sounds are captured more reliably. However, the microphone 2A, 2B, 2C may also be worn at other positions, for example as illustrated. The microphone 2A, 2B, 2C may be implemented as a stand-alone device. The microphone 2A, 2B, 2C may alternatively be integrated into a portable electronic device which also performs other functions, for example a device worn near the ear(s), such as a headphone, ear bud(s), in ear monitor(s), bone conduction headphones, hearing aids, etc. The microphone 2A, 2B, 2C may be integrated into the device itself or, as depicted by microphone 2A, into a connection cable connecting the device to another device of the person, for example a smart phone. The portable electronic device into which the microphone 2A, 2B, 2C is integrated may be, for example, a mobile communication device, such as a smart phone or smart watch.

[0123] In an embodiment where the microphone is implemented in a portable electronic device 4 configured to be worn on the wrist, the portable electronic device 4 may be implemented in the form of an electronic bracelet, electronic cuff, or electronic watch 4, for example. In particular, the wearable device can be implemented as a sports watch 4 (as depicted in Fig. 1), health tracker, fitness tracker, or sports computer.

[0124] The microphone 2A, 2B, 2C may be embedded, integrated or otherwise connected to an item worn or carried by the person 8, for example a garment, such as a t-shirt, hat, headband, neck warmer, etc.

[0125] The microphone 2A, 2B, 2C is, in some embodiments, not worn by the person 8 or directly carried by the person 8, but may be associated with the person 8 by virtue of the person 8 using equipment into which the microphone 2A, 2B, 2C is integrated, embedded or attached to. For example, the microphone 2A, 2B, 2C may be integrated or connected to a cycling computer attached to a bicycle the person 8 is riding, or integrated or connected to a treadmill the person 8 is running on.

[0126] Depending on the embodiment, the microphone 2A, 2B, 2C may be arranged in a device, such as a portable electronic device, which comprises further sensors, in particular sensors configured to measure a performance level and / or physiological values of the person 8, as described herein.

[0127] In an embodiment, rather than a microphone 2A, 2B, 2C, a more general transducer, the transducer designed to convert vibrations related to breathing into electrical signals. The vibrations may be vibrations of the air, but also vibrations of the body itself, as transmitted, for example, via tissues, e.g., the skeleton, of the person 8. These transducers include, for example, bone conduction microphones and digital stethoscopes.

[0128] The microphone 2A, 2B, 2C is connected to a processor (not depicted). The connection may be wired and / or wireless. The processor may be integrated into, or part of, the same device as the microphone 2A, 2B, 2C. The processor may, additionally or alternatively, be arranged separately.

[0129] Fig. 2 shows a block diagram which includes structural components of the electronic system 1 . The electronic system 1 comprises at least one processor 11 configured to carry out one or more steps and / or functions as described herein.

[0130] Depending on its configuration, the electronic system 1 further includes various components, such as a memory 12, a communication interface, and / or a human machine interface (HMI). The components of the electronic system 1 are connected to each other via a data communication system, such that they can transmit and / or receive data.

[0131] The term data communication system relates to a communication system that facilitates data communication between two components, devices, systems, or other entities. Depending on its configuration, the data communication system is wired and includes a wired connection, such as a cable and / or a system bus, and / or includes a wireless connection, such as Bluetooth (BT), Bluetooth Low Energy (BLE), ANT+, Wi-Fi, RFID, etc. The data communication system may further include communication modules for communication via networks, such as local area networks (LANs), mobile radio networks (e.g., GSM, GPRS, CDMA2000, EDGE, and / or UMTS), and / or the Internet 5. The Internet 5 includes, depending on the implementation, intermediary networks.

[0132] The processor 11 may comprise a system on a chip (SoC), a central processing unit (CPU), and / or other more specific processing units such as a graphical processing unit (GPU), application specific integrated circuits (ASICs), or reprogrammable processing units such as field programmable gate arrays (FPGAs).

[0133] The memory 12 comprises one or more volatile (transient) and / or non-volatile (nontransient) storage components. The storage components may be removable and / or nonremovable, and can also be integrated, in whole or in part with the processor 11. Examples of storage components include RAM (Random Access Memory), flash memory, hard disks, data memory, and / or other data stores. The memory 12 comprises a non-transitory computer-readable medium having stored thereon computer program code configured to control the processor 11 , such that the electronic system 1 performs one or more steps and / or functions as described herein. Depending on the embodiment, the computer program code is compiled or non-compiled program logic and / or machine code. As such, the electronic system 1 is configured to perform one or more steps and / or functions.

[0134] The computer program code defines and / or is part of a discrete software application. One skilled in the art will understand that the computer program code can also be distributed across a plurality of software applications (Apps). In an embodiment, the computer program code further provides interfaces, such as APIs, such that functionality and / or data of the electronic system 1 can be accessed remotely, such as via a client application or via a web browser. While particular steps and / or functions are described herein as being performed by a particular component or device of the electronic system 1 , particular steps and / or functions may be performed in other components or devices of the electronic system 1 in whole or in part. Further, particular steps disclosed as being performed by the processor 11 may be performed by the wearable device 2. Additionally or alternatively, particular steps as disclosed as being performed by the processor 11 may be performed by a plurality of processors 11 , in particular a plurality of processors 11 distributed across one or more devices of the electronic system 1.

[0135] The human machine interface (HMI) of the electronic system 1 includes input means, using which the person 8 may provide data to the electronic system. The input means include, for example, a touch interface, buttons, keys, a mouse or other tactile input mechanisms, a microphone (which may be the same microphone as used for capturing the breathing-related sounds), a camera (for gesture recognition, for example), etc. The HMI further includes output means for providing information, for example in the form of one or more of the messages described herein, to the person 8. The output means include means for visual, acoustic and / or haptic output, for example, a display, a speaker, or vibrational mechanism.

[0136] As depicted in Figs. 3 to 6, the electronic system 1 may be implemented in a variety of ways, using one or more devices connected to each other. These examples are not intended to be limiting, rather they are to illustrate some possible ways of implementing the electronic system 1 according to the invention.

[0137] In Fig. 3, the microphone 2 is connected to a portable electronic device 4, in this case a smart phone 4, of the person, the smart phone 4 having a HMI including at least a display 41 . The microphone 2 may be connected wirelessly to the smart phone 4 or connected using a wired connection. The smart phone 4 further comprises a processor (not shown) and memory (not shown). In Fig. 4, the portable electronic device 4 of the person is a smart watch, sports watch, or other wrist worn portable electronic device 4, the wearable device 4 having a HMI including at least a display 41 .

[0138] In an embodiment, the microphone 2 depicted in Figs 3. or 4 may be integrated into the portable electronic device 4, in particular the smart phone or the wearable device, respectively.

[0139] Analogously to the portable electronic device 4 worn on or carried directly by the person, the functional and / or structural features of the portable electronic device 4 are, in an embodiment, provided by or implemented in an electronic device integrated or connected to equipment the person is using, for example a bicycle or a treadmill. For example, the electronic device may be implemented as a bicycle computer or a treadmill computer, respectively. References herein and features described in relation to the portable electronic device 4 may also apply to a more general electronic device.

[0140] The portable electronic device 4 may include a communications module configured for data communication with other devices, in particular with other devices of the electronic system 1 , using the data communication system. Depending on the embodiment, the communications module may be configured for wired and / or wireless communication.

[0141] Depending on the embodiment, the portable electronic device 4 may also include further electronic components, in particular a power source such as a battery. Other electronic components include, for example, an HMI configured to receive user input and / or provide information to the user, for example comprising user input means such as a touch screen, buttons, a rotary wheel, etc. The HMI may provide information to the user by means of a display (e.g., a screen, a touch screen, and / or an AR display system), a loudspeaker, or haptic feedback (e.g. using vibrations). Depending on the embodiment, the portable electronic device 4 is implemented as a mobile device, for example a mobile radio phone (e.g., a smart phone running an iOS or Android operating system), a tablet computer, a laptop computer, a smart watch, a fitness watch, a sports computer (e.g., an electronic sports watch, a cycling computer, or a rowing computer). The portable electronic device 4 may include a display 41 , for example implemented as a touch screen. The portable electronic device 4 is configured to be connected to the microphone 2 either using a wired or wireless data communication system.

[0142] In an embodiment, the portable electronic device 4 further comprises modules configured to determine a current time, an orientation of the portable electronic device 4, a location of the portable electronic device 4 (e.g. using a global navigation satellite system (e.g., GPS) receiver, signal strengths of nearby WLAN access points, or signal strengths of nearby mobile radio transceivers), and / or ambient conditions, which ambient conditions comprise an air temperature, pressure and / or humidity.

[0143] Additionally, depending on the embodiment, the portable electronic device 4 comprises a processing unit separate from the processor 11. The processing unit is, for example, implemented as a microprocessor running program code (for example, embedded program code. As such, the processing unit may be configured to perform one or more steps and / or functions as described herein. The processing unit is connected to the microphone 2, and to other electronic components of the portable electronic device 4, including the battery, user interface, communication module, etc.

[0144] Additionally, depending on the embodiment, the portable electronic device 4 comprises a memory connected to the processing unit configured to record the microphone signal and / or an output of one or more modules or sensors of the wearable device 4. Depending on the embodiment, the microphone signal received in the portable electronic device 4 may be pre-processed (e.g., adjusted, corrected, filtered, statistically analyzed, summarized, and / or compressed) in the processing unit of the portable electronic device 4.

[0145] Multiple microphones 2 may be connected to the portable electronic device 4. In such an instance, the portable electronic device 4 may process multiple microphone signals for the multiple microphones 2 and generate a single microphone signal output. The multiple microphone signals may be processed to increase a signal to noise ratio, to isolate and / or remove particular acoustic features, and / or to spatially filter the audio input, in particular such that breathing related sounds are isolated.

[0146] Depending on the embodiment, the portable electronic device 4 further comprises, or is connected to one or more sensors configured to measure a performance level of the person, the one or more sensors including, for example, a heart rate monitor, a power, a speed, skin or core temperature sensor, and / or a blood lactate monitor. The sensors may be configured to measure a heart rate, a power output, a speed or pace, a skin and / or core temperature, and / or a blood lactate level, respectively.

[0147] In Fig. 5, a distributed implementation of the electronic system 1 is illustrated, in which the microphone 2 is connected to a distributed processing environment 7 which includes a portable electronic device 4, such a smart phone, communicatively connected to a server computer 6 via an intermediary network 5 which may include a mobile radio network and / or the Internet. The server computer 6 includes a processor 11 and memory 12 and may be located remotely from the microphone 2 and the portable electronic device 4, for example in a data processing facility such as a cloud computing center.

[0148] One or more of the functions and / or steps described herein may be performed, in whole or in part, in the portable electronic device 4 and other functions and / or steps described herein may be performed, in whole or in part, in the processor 11. In particular, the steps and / or functions described herein as being performed on the processor 11 may be performed either in the processor 11 or in the portable electronic device 4 (the processor 11 being implemented in part in the portable electronic device 4). Additionally, in the case that the processor 11 is implemented, at least in part, in the portable electronic device 4, the portable electronic device 4 may have a separate processing unit as described herein, with the functions and / or steps described herein being performed either in the processing unit of the portable electronic device 4, the processor 11 of the server computer 6, or in conjunction between the two.

[0149] As depicted in Fig. 6, the microphone 2 may be integrated into the portable electronic device 4. The processor 11 and the memory 12 may be implemented in the server computer 6. However, at least some of the steps and / or functions performed by the processor 11 may be performed, in whole or in part, in the portable electronic device 4 and / or in conjunction with the portable electronic device 4. For example, the portable electronic device 4 may perform pre-processing of the microphone signal(s). The portable electronic device 4 may merely forward data received from the portable electronic device 4, in particular as received from the microphone, to the server computer 6 (forwarding the data may include buffering, e.g., temporarily storing, the data). For example, the server computer 6 may perform data processing and transmit messages to the portable electronic device 4 based on the processed data. The portable electronic device 4 is connected to the server computer 5 via an intermediary network 5 which may include a mobile radio network and / or the Internet. To that end, the portable electronic device 4 includes a mobile radio transceiver, for example configured to communicate using one or more digital cellular technologies (e.g., GSM, GPRS, CDMA2000, EDGE, and / or UMTS).

[0150] The portable electronic device 4 may include or be connected to a sensor system. The sensor system may include a photo plethysmograph (PPG) sensor. The PPG sensor may be integrated in the same device as further sensors, for example in the portable electronic device 4, or be arranged separately from the other sensors. In particular, the PPG sensor may be implemented in a wrist worn device, such as a fitness tracker or smart watch. The PPG sensor may alternatively be arranged in an electronic ring worn on a finger. The PPG sensor is configured to communicate, using the data communication system as described herein, depending on the embodiment, with the portable electronic device 4 and / or with the processor 11.

[0151] The sensor system may further include an ECG sensor. The ECG sensor may be integrated in the same device as the other sensors of the sensor system, for example in the portable electronic device 2, or be arranged separately from the other sensors. In particular, the ECG sensor may be implemented in a second wearable device worn in contact with the torso, in particular the chest, of the human. The second wearable device may be attached to the human by way of a strap or belt. The ECG sensor is configured to communicate, using the data communication system as described herein, depending on the embodiment, with the portable electronic device 4, and / or with the processor 11.

[0152] Depending on the embodiment, the sensor system may be configured to determine additional physiological signals using one or more sensors, including a galvanic skin response, a sweat rate, an oxygen saturation, a blood oxygen saturation, a blood pressure, a glucose concentration, or a lactate concentration.

[0153] The sensor system may further be configured to measure other values, including environmental conditions, such as an ambient air temperature using an ambient temperature sensor, and an ambient humidity, or values related to a location or movement of the human. The location may be determined using a GNSS receiver (e.g. a GPS receiver). The movement may be determined by means of an inertial measurement unit (IMU) comprising, for example, a three axis accelerometer, and / or gyroscope. Figs. 7 to 15 show flow diagrams illustrating methods 110, 120, 130, 140, 150, 160, 170, 180, each comprising a number of steps performed by the electronic system 1 , in particular performed by the portable electronic device 2 and / or the processor 11 . These methods 110-180 may be performed, depending on the particular method, prior to, during, or after exercise. The methods may be performed based on recorded microphone signals and / or microphone signals received contemporaneously with their capture (e.g., in real-time). The methods 110-180 may be performed once, in an intermittent fashion, or in a continuous fashion. The methods 110-180 or particular steps thereof may be performed on demand, i.e. when an appropriate signal is received from a user or from a device which triggers the method or steps, or the methods or particular steps thereof may be performed automatically, for example at pre-determined times or after predetermined periods.

[0154] Fig. 7 shows a flow diagram illustrating a method 100 which includes a number of steps S100-S104. The method 100 may be performed on demand, for example in response to user input, or automatically, for example upon the processor 11 receiving an indication that the person 8 has begun exercising. The method 100, or particular steps thereof, may be performed as part of other methods described herein.

[0155] In step S100, the microphone 2 receives an acoustic signal, for example via the air (but could also be via water, if the person is swimming, for example). The acoustic signal includes breathing sounds from the person.

[0156] In step S101 , the microphone 2 generates a microphone signal according to the acoustic signal. The microphone signal may be an analog or digital signal.

[0157] In step S102, the microphone 2 transmits the microphone signal, in a transmission T1 , for example via a wired and / or wireless connection. The wired connection includes, for example, a cable, a trace, or a system bus. The wireless connection includes, for example, a short range radio transmission such as a BT, BLE, ANT+, or UWB transmission. Transmitting the microphone signal may include using a communication interface, such as a wireless radio.

[0158] In step S103, the processor 11 receives the transmission T1 of the microphone signal.

[0159] In step S104, the processor 11 processes the microphone signal, in particular as described in one of the methods disclosed herein. For example, the processor 11 generates a breathing rate BR of the person using the microphone signal. Alternatively or additionally, the processor 11 generates a ventilation volume of the person using the microphone signal.

[0160] The processor 11 may also store the microphone signals received over a particular period of time in the memory to generate a recorded microphone signal. For example, the processor 11 may store the microphone signals received during an initialization phase, where the person is at rest and breathes normally, preferably in a quiet environment. In another example, the processor 11 may store the microphone signals received during a test phase, in which the person performs exercise according to a predefined protocol, for later analysis.

[0161] The steps S100 - S104 are typically performed in such a fashion that current acoustic signals received by the microphone are being constantly used to generate microphone signals which are transmitted to the processor 11.

[0162] Fig. 8 shows a flow diagram illustrating a method 110 which includes a number of steps S111 - S113. The method 110 is performed in the electronic system 1, in particular in the processor 11. The method 110 is performed to determine one or more ventilatory thresholds of the person. As such, the method 110 may be performed only once for a particular person or whenever desired by the person. For example, if their fitness changes or they have a change in their health status, the person may wish to perform method 110 anew.

[0163] In step S111 , a recording of a microphone signal is received. The recorded microphone signal is a recording of the microphone signal over a period of time, in particular as recorded by the microphone during a test phase.

[0164] In an embodiment, the microphone signal is enhanced to improve audibility of breathing. For example, a high pass filter may be applied to filter out ambient sound, specifically wind noise and / or traffic noise. The high pass filter may have a frequency of between 500 Hz to 1.5 kHz, preferably 1 kHz, as the majority of wind noise occurs below approximately 800 Hz. Breathing related sounds above 1 kHz thereby remain.

[0165] In an embodiment, a spectrogram is generated from the microphone signal, for example as shown and described with reference to Fig. 17), and the spectrogram used for subsequent processing and analysis of the microphone signal. The spectrogram may be normalized, for example by scaling it from 0 to 1 .

[0166] In an embodiment, the person performs specific exercise during the test phase, in particular according to a pre-defined exercise protocol such as a ramp test. During the ramp test, for example, the breathing rate of the person will tend to steadily increase until (or until shortly after) the person fatigues (also known as reaching volitional exhaustion) and is unable to continue exercise.

[0167] Preferably, the pre-defined exercise protocol includes at least two periods of different exercise intensities. One of the exercise intensities may be “rest”, i.e. no intensity at all. One of the exercise intensities may be an “active recovery” intensity. During rest or active recovery, the breathing rate is expected to be at rest or at a baseline value. The breathing rate at rest, and therefore the baseline value, is typically in the range of 10 to 20 breaths per minute.

[0168] In an example, the pre-defined exercise protocol is designed to elicit, in the person, a maximum breathing rate. The maximum breathing rate is typically in the range of between 30 and 60 breaths per minute.

[0169] The pre-defined exercise protocol may be provided to the processor, for example via user input, or the processor may be configured to analyze the breathing rate or other received signals from sensors to determine the particular exercise protocol used. For example, a received power from a power meter or other ergometer, or a received running speed or running pace, is typically sufficient information for the processor to deduce the particular exercise protocol used.

[0170] In step S112, breathing rate values are determined using the microphone signal. In particular, a plurality of breathing rates are determined from the microphone signal according to the breaths identified in the microphone signal.

[0171] As described herein, the breaths, and consequently the breathing rate, may be determined using a breathing rate analysis module, also referred to as a breathing analysis model.

[0172] Optionally, prior to determining the breathing rate values, pre-processing of the microphone signal may take place as described herein, for example to remove noise or artifacts due to steps or other movement of the person.

[0173] In step S113, the breathing rate values are analyzed to determine one or more ventilatory thresholds of the person. In particular, one or two ventilatory thresholds may be identified. The ventilatory thresholds may be determined from the breathing rate values using a rules based approach or a data driven approach. In an embodiment, a rules based approach is implemented, for example, to determine a particular ventilatory threshold as a defined absolute or relative difference between the resting breathing rate and / or the maximal breathing rate. For example, the first ventilatory threshold may be defined as being between 3 and 15 breaths per minute above a baseline breathing rate, or between 40% and 80% higher than the baseline breathing rate. The second ventilatory threshold may be defined as being between 16 and 30 breaths per minute above the baseline breathing rate, or between 230% and 400% of the baseline breathing rate.

[0174] In an embodiment, a data driven based approach was implemented as follows. A ventilatory threshold dataset including breathing rate values and manually labeled first and second ventilatory thresholds on the basis of CPET was used. The ventilatory threshold dataset had a size of several hundred people, including male and female, having a large age range, and a training subset of the ventilatory threshold dataset was used to train a ventilatory threshold model, in particular a decision tree, more particularly, an AdaBoost optimized decision tree, to predict the labeled first and second ventilatory thresholds based on the breathing rate alone. The trained ventilatory threshold model was then tested on a test subset of the ventilatory threshold dataset and the mean absolute error and mean absolute percentage error of the trained ventilatory threshold model was 2.17 and 8.72%, respectively. The trained ventilatory threshold model may be used to determine, from breathing rate values determined according to step S112, both the first and the second ventilatory threshold of the person.

[0175] In optional subsequent steps, the ventilatory threshold(s) may be stored in the memory, transmitted to an external device, and / or provided to the person. For example, the ventilatory thresholds may be displayed to the person via the HMI. Thereby, the person is informed of their ventilatory thresholds and may, for example, design subsequent exercise, e.g., during training or competition, using the ventilatory thresholds to optimize training and / or performance. Fig. 9 shows a flow diagram illustrating a method 120 which includes a number of steps S121 - S123. The method 120 associates a resistance level provided by exercise equipment and / or an intensity level of exercise with the one or more ventilatory thresholds. This method 120 may be performed subsequent to method 110 in which the ventilatory thresholds are determined. Alternatively, the method 110 may be performed as part of the method 120.

[0176] In step S121 , a control signal is generated for the exercise equipment. The control signal is preferably generated based on a pre-defined exercise protocol (e.g., a ramp test, or a specific set of intervals). The control signal is configured to control a resistance level of the exercise equipment. For example, in the case of a treadmill, the control signal may define a speed and / or an inclination of the treadmill over time, according to the predefined exercise protocol.

[0177] In step S122, the control signal is transmitted to the exercise equipment, for example via a communication interface.

[0178] Steps S111 , S112 and S113 are then performed, as described with reference to method 110.

[0179] In step S123, the resistance level and / or the exercise intensity are associated with the one or more determined ventilatory threshold(s). In particular, the resistance level at or around a time-point at which the ventilatory threshold was reached is associated with that particular ventilatory threshold. This is possible because the control signal, or the transmission of the control signals to the exercise equipment, defines the resistance level at particular time-points and the breathing rate at those particular time-points is also determined. For example, if the first ventilatory threshold was determined to occur at a breathing rate of 18 breaths per minute, and this breathing rate was reached when the treadmill was providing a resistance level of a 5 min / km pace at a 0% inclination, then the first ventilatory threshold is associated with a resistance level of 5 min / km. Similarly, if the second ventilatory threshold was determined to occur at a breathing rate of 28 breaths per minute and this occurred at a resistance level of 4:10 min / km, then the second ventilatory threshold is associated with a resistance level of 4:10 min / km. Similar considerations apply to associating the exercise intensity (objectively or subjectively determined) with the ventilatory threshold(s).

[0180] Thereby, the person may subsequently train relative to a particular ventilatory threshold without monitoring their breathing. In particular, they may set their exercise equipment to a particular resistance level, or train at a particular defined rating of perceived exertion relative to a particular ventilatory threshold. For example, they may wish to train just below the first ventilatory threshold, or train just above the second ventilatory threshold, such as to elicit a particular training stimulus.

[0181] Fig. 10 shows a flow diagram illustrating a method 130 which includes a number of steps S131 - S132. The method 130 associates a performance level provided by sensors with the one or more ventilatory thresholds. This method 130 may be performed subsequent or parallel to the method 110 in which the ventilatory thresholds are determined. Alternatively, the method 130 may be performed at the same time, or subsequent to, the method 120.

[0182] Steps S111 , S112 and S113 are performed as described herein with reference to method 110.

[0183] In step S131 , a performance signal is received from a sensor or from a sensor system. The performance signal is indicative of a performance level of the person during the exercise, and in particular provides the performance level of the person at particular breathing rates. The performance signal may be indicative, for example, of a power output, pace, speed, heart rate, skin and / or core temperature. The performance signal is indicative of the

[0184] In step S132, the one or more ventilatory thresholds determined in step S113 are associated with one or more performance levels, respectively. For example, the first ventilatory threshold may be determined to occur at a heart rate of 135 beats per minute for the particular person. The second ventilatory threshold may be determined to occur at a heart rate of 165 beats per minute, for example.

[0185] Armed with this knowledge, the person may then subsequently receive, based on their monitored heart rate alone (for example as provided by a sports watch with a heart rate monitor), an indication of which training zone they are in, without having to directly or indirectly monitor their breathing, in particular their breathing rate.

[0186] Fig. 11 shows a flow diagram illustrating a method 140 for determining a ventilatory zone, comprising steps S141-S143. The method 140 may be performed independently from other methods described herein, in particular independent of methods 110-130. The method may be performed during exercise, or after exercise. In other words, the method may be performed using microphone signals received in real-time, or in near real-time, or may be performed based on recorded microphone signals which were recorded during exercise. The former allows for providing the person with a real-time indication of their ventilatory zone, the latter for analysis of their exercise after the exercise has been completed.

[0187] In step S141 , a microphone signal is received. The microphone signal is received, for example, as described with reference to method 100 and steps S100-S103. In step S142, the breathing rate of the person is determined. The breathing rate determined may be a current breathing rate, or one or more prior breathing rate values. The breathing rate may be determined using a breathing analysis model as described herein.

[0188] In an optional step, the ventilation volume of the person is determined. The ventilation volume of the person may be determined using the breathing analysis model as described herein.

[0189] In step S143, the ventilatory zone of the person is determined using the breathing rate and one or more pre-defined ventilatory thresholds. The ventilatory thresholds may be defined as a function of the breathing rate, and the pre-defined ventilatory threshold(s) may have a default value, a value defined based on personal information of the user, and / or individual values determined using the method 110 described herein.

[0190] The ventilatory thresholds separate the ventilatory zones. For example, the first ventilatory threshold separates the first and second ventilatory zones, and the second ventilatory threshold separates the second and third ventilatory zones.

[0191] In the case that the method 140 is analyzing a recorded microphone signal, recorded during exercise, the ventilatory zone of one or more periods within the period of exercise may be determined. For example, it may be determined that for the first half of their exercise, the person was in the second ventilatory zone, and for the second half, they were in the first ventilatory zone.

[0192] For example, if the current breathing rate is 15 breaths per minute, and the first ventilatory zone is 18 breaths per minute, then the person is determined to be in the first ventilatory zone. The determined ventilatory zone may be stored in memory, transmitted to a device, and / or provided to the person via the HMI. Thereby, the person is informed of the ventilatory zone, e.g. their current ventilatory zone, or a ventilatory zone they were in at a particular point in time during their exercise. Thereby, they are provided with useful information related to their current or past physiological state.

[0193] Fig. 12 shows a flow diagram illustrating a method 150 for determining a breathing rate of a person. The method 150 includes steps S151 and S152 and may be performed as part of step S142 as described with reference to method 140. This method 150 describes one way of analyzing the microphone signal to determine the breathing rate. The method 150 may be implemented as part of a breathing analysis model as described herein.

[0194] In step S151 , a single breath is identified in a part of the microphone signal. In particular, a defined time-window of the microphone signal, (i.e. a snippet of the microphone signal), which time-window may have a length of between 0.1 s and 3 s, is analyzed to determine whether a breath (optionally, two or more breaths) has occurred within the time-window or is taking place in the time-window. The length of the time-window may be dynamically adjusted according to a current (or past) breathing rate. In particular, the length of the time-window may be increased if the breathing rate decreases, and the length of the time-window is decreased if the breathing rate increases. Thereby, it is likely that only one breath will occur in the time-window providing for a more reliable breath identification.

[0195] The time-window may have an associated time-point, which time-point identifies the part of the microphone signal, in particular using a relative or absolute time-point at which the time-window is arranged. The time-point may define, for example, a beginning, middle, and / or end of the time-window. In an example, a pre-defined acoustic fingerprint may be used as described herein, along with pattern matching, to identify a breath or a part thereof. If a breath is identified, then the particular time-point at which the breath was identified may be marked, or otherwise associated with a breath, such as to avoid double-counting a particular breath. If a breath is identified, a current breathing rate may be adapted or updated accordingly, and / or a message may be generated indicative of an identified breath.

[0196] In an embodiment, a match score is determined for each identified breath, the match score indicative of a probability that the breath was correctly identified. The match score may be determined using pattern matching. The breath is identified, i.e. used to determine the breathing rate, only if the match score satisfies one or more pre-defined thresholds.

[0197] In an embodiment, the breathing analysis model is configured to identify, in the microphone signal, an exhalation and to determine the breathing rate using a length of time between two or more exhalations. Additionally or alternatively, the breathing analysis model may be configured to identify, in the microphone signal, an inhalation and to determine the breathing rate using a length of time between two or more inhalations.

[0198] In an embodiment, each time-window is labeled as including either an exhalation or an inhalation. In particular, the breathing analysis model may be configured to determine a particular label for a particular time-window by matching, using pattern matching, one or more pre-defined acoustic fingerprints with the time-window.

[0199] Step S151 is performed repeatedly across the microphone signal to identify all breaths in the microphone signal. The time-windows may be overlapping or non-overlapping. In an embodiment where the microphone signal is a “live” signal, i.e. being received in realtime or near real-time from the microphone, step S151 may be performed repeatedly for a time-window which includes the most recently received microphone signal, i.e. performed every second for a time-window which includes the latest 1 - 3 seconds of microphone signal. In step S152, the breathing rate is determined using a sequence of time-windows, in particular using a number of unique breaths identified in sequence of time-windows.

[0200] In an embodiment, the breathing rate, in particular a current breathing rate or a moving average current breathing rate) is used to determine a time-point of a time-window. For example, if the breathing rate is currently, or at a particular point in time, 15 breaths per minute (i.e., a breath every four seconds), the time-point of a next, future, or subsequent time-window may be determined to be four seconds after a time-point of a last breath. The breathing rate, in particular time-points of past breaths, is thereby used to predict, by extrapolation time-points or time-ranges within which future breaths are likely to occur. This is reliably possible because the breathing rate tends to change slowly.

[0201] In an embodiment, the match score is further determined using a predicted time-point of a future breath. In particular, the method includes generating a higher match score if the breath occurs at, or within a defined range of, the predicted time-point of the future breath.

[0202] Optionally, prior to step S151 , the microphone signal may be filtered using a digital filter as described herein, in particular to reduce environmental noise, and / or noise from footsteps or other movements of the person.

[0203] Fig. 13 shows a flow diagram illustrating a method 160 for generating a training dataset suitable for training a breathing analysis model. The method 160 includes steps S161 - S165. The method 160 further includes optional steps S166 and S167. The method 160 needs to be performed only once to obtain the training dataset. Thereafter, a breathing analysis model may be trained using the training dataset, in particular using machine learning. In an embodiment, the method 160 does not include step S165, i.e. the training dataset is not stored. Rather, the labeled microphone signal generated in S164 and optionally in S167 is used directly for training (in particular, for updating) a breathing analysis model and is then discarded.

[0204] In step S161 , a microphone signal is received from a person. The microphone signal is received from a microphone associated with the person, preferably worn on the person as described herein. The person may be instructed, for example by instructions generated by the processor and displayed or otherwise provided to the user, to breathe according to a defined breathing protocol and / or to exercise according to a pre-defined exercise protocol (e.g., ramp protocol) as described herein. The breathing protocol may define one or more breathing rates (total breathing rate, speed of inhalation, speed of exhalation, time between inhalation and exhalation, and / or time between exhalation and inhalation).

[0205] The breathing protocol may further define one or more breathing depths and / or breathing models, e.g., shallow breaths and / or deep breaths. In particular, the breathing protocol includes relaxed deep breathing (however must be audible), simulated exhausted breathing (hyperventilation), wheezing (simulating an asthma attack), and / or exhausted breathing (recorded directly after an intensive exercise session).

[0206] The microphone signal may be pre-processed, in particular using a Butterworth (low- pass) filter to remove high frequencies. Low frequency noise due to wind or body movement artifacts may also be removed, in particular by filtering our signals below 5 Hz.

[0207] Simultaneously, step S162 is performed, in which a dynamic chest expansion signal is received. Thereby, the microphone signal and dynamic chest expansion signals may be associated with each other. The microphone signal and dynamic chest expansion signals may be synchronized with each other. For example, a timing signal from a reliable clock is used. The timing signal may originate from a device recording the microphone signal (e.g., a smart-phone connected to which the microphone is connected), a device recording the dynamic chest expansion signal (e.g., the electronic chest harness described below), or a further device. The timing signal is associated with both the received dynamic chest expansion signal and the microphone signal during recording of the dynamic chest expansion signal and the microphone signal, such that both may be recorded in a synchronized manner, e.g., with time-stamps which correspond to each other correctly.

[0208] In addition or alternatively to the dynamic chest expansion signal, additional breathing- related physiological data of the person may be received, for example a flow meter signal is received from a flow meter configured to measure the breathing of the person directly, the flow meter signal indicative of a breathing rate and / or a ventilation volume.

[0209] The dynamic chest expansion signals are received from a purpose built electronic chest harness which includes an elastic strap arranged around the torso such that the elastic strap stretches during inhalation. Sensors on the electronic chest harness are configured to measure the elongation or stretching of the elastic strap over time record the dynamic chest expansion signal. The electronic chest harness is configured to transmit the dynamic chest expansion signals to the processor, for example. In particular, a magnet is placed in a fixed position on the chest harness and a flux sensor is arranged on the elastic strap such that the distance between the flux sensor and the magnet varies based on expansion and contraction of the chest. The flux sensor is configured to determine the magnetic field of the magnet and, due to the 1 / rA4 relationship between the magnetic flux and its distance from the magnet, this arrangement provides highly accurate measurements of the chest circumference during breathing. The electronic chest harness may be further configured to measure a bio-impedance signal, and to transmit the bio-impedance signals, which are also indicative of breathing, to the processor.

[0210] The electronic chest harness may further comprise an I MU, a three-axis magnetometer and / or a temperature sensor to collect further data. In particular, the IMU may be used to collect data for identifying steps or other movements of the person, such that the labels, indicative of breathing and / or steps, can be more accurately applied.

[0211] The dynamic chest expansion signals may be smoothed, for example using a smoothing filter such as a Savitzky-Golay filter to remove motion artifacts while preserving signal peaks.

[0212] The dynamic chest expansion signal and / or the microphone signal may further be pre- processed by interpolating missing data points, for example due to temporary signal loss in the case of wireless signal transmission. The dynamic chest expansion signal and / or the microphone signal may further be resampled, for example such that both signals are sampled in the same or a corresponding frequency. The dynamic chest expansion signal and / or the microphone signal may further be normalized, for example by scaling their values from between 0 to 1.

[0213] In step S163, ground truth breath time-points are determined using the dynamic chest expansion signal. In particular, an onset of expansion, i.e. the particular time-point when the elastic begins to stretch from a minimally stretched configuration, is indicative of inhalation. The onset of contraction, i.e. the time-point when the elastic begins to contract from a maximally stretched configuration, is indicative of exhalation. Further, a duration of stretching and / or a degree of elongation, is indicative of a ventilation volume. In an embodiment, the ground truth breath time-points may be determined by extracting peaks from the data, e.g. such that a peak amplitude of the dynamic chest expansion signal may be considered to correspond to the time-point of a breath.

[0214] The ground truth breath time-points may further be determined using the bio-impedance signals.

[0215] In step S164, the breath time-points, as determined using the time of onset of expansion and / or the time of onset of contraction, are used to generate a labeled microphone signal in which the labels are indicative of breathing time-points (and / or may be indicative of time-points of an inhalation and / or an exhalation).

[0216] In an embodiment, in addition or alternatively to labeling the microphone signal with breath time-points, the microphone signal may be labeled at a plurality of time-points with the breathing rate at that time-point (e.g,. a breathing rate time-series is generated and associated with the microphone signal).

[0217] A median filter may be applied to the breathing rate time-series such that the breathing rate does not vary too quickly, which would be the case if the breathing rate is determined at a high level of granularity, e.g., by calculating the breathing rate anew for every inhalation). For example, the median filter may provide a median breathing rate smoothed over 3 to 10 seconds, preferably 5 seconds.

[0218] In step S166, the dynamic chest expansion signal is used to determine ground truth ventilation volumes associated with one or more breaths. The ventilation volume may be based on the relative expansion of the chest harness during breathing. The ground truth ventilation volume may be a relative ventilation volume and / or an absolute ventilation volume. Additionally or alternatively, the flow meter signal is used to determine the ground truth ventilation volumes associated with one or more breaths, in particular in conjunction with the pre-defined exercise protocol and / or breathing protocol.

[0219] In step S167, labels are applied to the microphone signal indicative of the ventilation volumes associated with one or more breaths.

[0220] Optionally, ground truth labels for other sound events, may be generated and associated with their appropriate time-points in the microphone signal. Other sound events associated with the person include, for example, steps or other motions, and coughing, wheezing, pulling up nose, etc. In an embodiment, in addition to the labels generated using the dynamic chest expansion signal and / or the flow meter signal, the microphone signals may be analyzed using a pre-defined breath signature designed to analyze the microphone signals of the training dataset. The pre-defined breath signature may have been generated previously using one or more particularly clean recordings of breathing, in for example recordings from a low noise environment recorded using more sophisticated recording microphones. The present pre-defined breath signature may be generated by extracting features from the time-series amplitude data and / or from the spectrogram (generated using short-time fourier-transform with 40 ms frames and 75% overlap). In an example, approximately 20-40 mel-frequency cepstral coefficients (MFCC) are extracted directly from the amplitude data to form the features. Additionally or alternatively, a large number of MFCC are extracted from many different breath samples and then a dimensionality reduction method such as principal component analysis (PCA) is applied, with the first 5-10 principal components forming the features. Additionally or alternatively, a large number of spectrograms from different breathing samples are created, PCA performed, in the frequency space to extract frequencies that change the most between breaths. These frequencies are then discarded, the remaining frequencies being those which are consistently present in multiple breaths. These frequencies then form the features. Additionally or alternatively, recursive feature elimination is used on cepstral coefficients to select those forming the features. Additionally or alternatively, a fisher score is used to score the cepstral coefficients, the highest scoring forming the features.

[0221] The pre-defined breath signature is then used to detect, in the recorded microphone signals, breaths, for example in a similar manner how the pre-defined acoustic fingerprint is used to identify breaths as described herein. In particular, the recorded microphone signal is divided into short audio snippets (e.g., 0.1 - 5s length) and features extracted corresponding to those features of the pre-defined breath signature. A similarity metric is then applied to determine if a breath is included in the audio snippet. The breaths detected using the pre-defined breath signature are then compared to those identified using the dynamic chest expansion signal and / or the flow meter signal. Any discrepancies may be manually checked and corrected, with correct labels applied.

[0222] Steps S161 to S164, and optionally steps S166 and S167, are performed either repeatedly during exercise, such that a labeled microphone signal is generated in realtime or in near real-time, or performed after exercise using a recorded microphone signal and a recorded dynamic chest expansion signal. The steps S161 and S165 are performed for a large number of people, preferably of different age, sex, and cardiovascular ability, such that the training dataset is representative of a large proportion of the population.

[0223] In step S165, the training dataset is stored in memory. The training dataset may further indicate, for each microphone signal, details such as the exercise protocol and / or breathing protocol and / or details related to the person. Thereby, the training dataset may be used to train a plurality of breathing analysis models specific for particular groups of people or an exercise type, for example. The details related to the person may include, for example, gender, age, height, and / or weight.

[0224] Fig. 14 shows a flow diagram illustrating a method 170 for training a breathing analysis model, including steps S171-S173. The method 170 is performed on demand for training a breathing analysis model. The breathing analysis model may be exercise type (e.g., sport) specific and / or specific to a group of persons.

[0225] Any discrepancies may be manually checked and corrected, with correct labels applied. The method 170, more particularly, the specific step S173 may be performed immediately subsequent to steps S161 - 164 (and optionally steps S166 and S167) of method 160 above. Thereby, the step of storing the training dataset and retrieving the training dataset is skipped, and the breathing analysis model is trained (or updated) without a need to store the data.

[0226] In step S171 , the training dataset, in particular as generated according to method 160, is received, for example retrieved from memory.

[0227] Optionally, the labeled microphone signals included in the training dataset are filtered in that labeled microphone signals with high-quality breathing signals (i.e. relatively high signal to noise ratio) are kept and other microphone signals, in particular those with strong background noise (for example due to wind), or where the microphone signal is saturated and / or clipped, discarded.

[0228] Optionally, noise reduction is applied to the labeled microphone signals, for example using spectral grating, to improve the signal to noise ratio of the microphone signals, in particular those filtered as being noisy.

[0229] In step S172, a breathing analysis model is initialized. The breathing analysis model may include a random forest model and / or a neural network. The neural network may include one or more fully connected layers, convolutional layers, attention layers, or a combination thereof.

[0230] The breathing analysis model may include a logistic regression model.

[0231] In step S173, the breathing analysis model is trained, using machine learning, in particular supervised machine learning, using the training dataset. In particular, the breathing analysis model is trained to correctly identify breathing time-points in the microphone signals, the training dataset including labels which serve to provide feedback, in particular in the form of an error generated by an error function, as to whether the breathing time-points were correctly identified. A part of the training database may be set aside for testing and / or validation of the model.

[0232] For example, the breathing analysis model is implemented as a neural network (e.g., a convolutional neural network, recurrent neural network, general adversarial network, variational autoencoder, or attention based network (e.g., transformer), or another type of neural network suitable to be trained using supervised learning, using an appropriate loss function and regularization.

[0233] An exemplary implementation of the breathing analysis model implemented as a neural network is shown and described with reference to Fig. 18 herein.

[0234] For example, the microphone signals may be divided into short audio-snippets, i.e. timewindows (which may overlap each other) of a defined time-length, for example between 0.1 s and 30 s, preferably between 0.1 s and 10 s, preferably between 0.1 s and 5 s, or preferably between 0.5 s and 10 s. The audio-snippets may include a breath or not. The audio-snippets may specifically include zero, one, or multiple breaths. The breathing analysis model learns, using the labels and supervised learning, to determine whether a given audio-snippet includes a breath and optionally at which particular time-point within the audio-snippet the breath occurs. The breathing analysis model may be configured to determine specific types and / or parts of a breath, for example an exhalation and / or an inhalation, or the beginning, middle, and / or end of an inhalation and / or an exhalation, respectively. The breathing analysis model may be configured to learn, using the labeled ventilation volume, a ventilation volume associated with a particular breath and / or part of a breath.

[0235] The breathing analysis model may be configured to determine whether the given audiosnippet includes zero, one, or more breaths, and optionally the time-point(s) at which the breath(s) occur. Using the number of breaths in the audio-snippet and / or the timepoints) at which they occur, the breathing analysis model may calculate or otherwise provide an indicator of the breathing rate.

[0236] The breathing analysis model may be configured to detect other sound events, in particular other sound events associated with the person, such as steps or other motions, and coughing, wheezing, pulling up nose, etc.

[0237] The breathing analysis model may generate an acoustic fingerprint associated with a breath and / or parts of a breath as described herein. The acoustic fingerprint for a breath, part of a breath, or other sound event, may, for example, be generated by averaging the signal, either in the time-domain and / or the frequency domain, for each time-snippet which has an appropriate label.

[0238] In an embodiment, the trained and / or updated breathing analysis model is transmitted to one or more user devices 4.

[0239] Fig. 15 shows a flow diagram illustrating a method 180 for determining a performance zone using the breathing rate and heart rate. The method comprises steps S181-S183. The performance zone, which may be defined, for example, by a three zone model, a six zone model, or a seven zone model, provides a sequence of zones of increasing exercise intensity. Neither the breathing rate nor the heart rate, for example, provide a sufficiently complete assessment of a current physiological state of the person as it relates to exercise performance. Therefore, according to the method 180, the breathing rate and the heart rate are combined to determine a performance zone.

[0240] In step S181, the breathing rate of a person is received, the breathing rate having been determined, for example, as described herein using a microphone signal.

[0241] In step S182, the heart rate of a person is received from a heart rate monitor worn on or by the person.

[0242] In step S183, a performance zone of the person is determined using the breathing rate and the heart rate.

[0243] Fig. 16 shows two charts in which the breathing rate and the heart rate, respectively, are plotted as a function of time and power during a ramp test. The ramp test is performed by a person on a stationary bicycle ergometer, in which the person must overcome the resistance level defined by a power in order to continue cycling. The ramp test is a predefined exercise protocol during which the resistance level increases steadily or step- wise until the person cannot continue.

[0244] Both charts relate to data collected during the same ramp test, the top chart showing the breathing rate during the ramp test and the bottom chart showing the heart rate during the ramp test. The ventilatory thresholds VT1 and VT2 are indicated with dashed and dotted vertical lines, and the ventilatory zones Z1, Z2, Z3 are also shown relative to the ventilatory thresholds VT 1 , VT2.

[0245] The power during the ramp test begins at 0 watts and remains there for 3 minutes. The breathing rate is at a baseline breathing rate of approximately 11 breaths per minute, though there are some fluctuations. The heart rate is at a baseline heart rate of approximately 63 beats per minute. The person is below the first ventilatory threshold VT 1 and consequently in ventilatory zone Z1.

[0246] Just before three minutes have elapsed, the power rises to 50 watts, marking a warmup phase with constant power that lasts another 3 minutes. At six minutes into the test, after the warm-up phase concludes, the power is increased steadily, while the breathing rate fluctuates around a relatively constant value of about 18 breaths per minute until around 14 minutes, i.e. at a power of 250 watts. This point marks the first ventilatory threshold VT1. The person has therefore transitioned across the first ventilatory threshold VT1 and is now in ventilatory zone Z2. In the meantime, the heart rate has increased to approximately 146 beats per minute. As the power climbs beyond VT 1 , the breathing rate increases approximately linearly until VT2 is reached shortly after 17 minutes into the test at a power of around 331 watts and a breathing rate of around 27 breaths per minute. The heart rate at this time-point is at 169 beats per minute. The breathing rate at powers beyond VT2 rises in an almost exponential manner and continues to climb until the person reaches volitional exhaustion and is unable to continue, the breathing rate reaching a maximum value of 44 breaths per minute.

[0247] Fig. 17 shows two charts of the same microphone signal recording, the top chart showing the relative signal amplitude and the bottom chart showing the spectrogram. The time axis is the same for both charts. The charts show a sample of a microphone signal of a duration of approximately ten seconds of a person performing exercise (in this case, running). The top chart further shows, as circles, time-points of steps ST. As can be seen, the rhythm of the steps is regular and the cadence is approximately 156 steps per minute.

[0248] A breathing analysis model as described herein was used to analyze the microphone signal to determine the breathing rate. The breathing analysis model was used with a time-window having a length of 0.3 s. An overlapping sequence of time-windows was applied to the microphone signal and, in each time-window, it was identified whether an exhalation EX was taking place or an inhalation IN.

[0249] Every time-window in which an exhalation EX was identified is marked with an ‘x’. In particular, the breathing analysis model uses a pre-defined acoustic fingerprint of an exhalation to identify whether a particular time-window of length 0.3 s includes an exhalation EX.

[0250] The pre-defined acoustic fingerprint of the exhalation may comprise a spectrogram, i.e. a specific set of frequencies and / or frequency ranges, along with a particular set of amplitudes and / or relative amplitudes between the frequencies and / or frequency ranges. The spectrogram of the pre-defined acoustic fingerprint of an exhalation is compared, using pattern matching, with the spectrogram of the particular time-window. In particular, using pattern matching, the breathing analysis model computes a match score indicative of a degree of correspondence between the spectrogram of the pre-defined acoustic fingerprint and the spectrogram of the particular time-window. If the match score satisfies a pre-determined match score threshold, then the time-window is identified as including an exhalation EX. For example, the spectrogram of the pre-defined acoustic fingerprint of the exhalation includes a particular set of frequency ranges. For a particular timewindow to be identified and / or labeled as including an exhalation EX, the spectrogram of the particular time-window must include frequencies in at least 30%, preferably at least 50% of the frequency ranges of the pre-defined acoustic fingerprint of the exhalation.

[0251] As can be seen, a large number of time-windows between 0 and 1.5 seconds identified an exhalation EX taking place. However, the breathing analysis model is configured to identify that only a single exhalation took place between 0 and 1.5 seconds, in particular because, during this time-window, no inhalation IN was detected. Analogously, every time-window in which an inhalation IN was identified is marked with a triangle. More specifically, the breathing analysis model uses a pre-defined acoustic fingerprint of an inhalation to identify whether a particular time-window of length 0.3 s includes an inhalation IN. As described above with reference to the exhalation EX, the breathing analysis model computes a match score by using pattern matching between the pre-defined acoustic fingerprint of the inhalation and a particular time-window to identify an inhalation IN in that particular time-window. The breathing analysis model only counts a single inhalation IN between approximately 1.4s and 2.25 s, because no exhalation EX took place during this time.

[0252] The breathing analysis model identifies sequences of time-windows including an exhalation EX and sequences of time-windows including inhalations IN. A breath is identified once a sequence of time-windows including an exhalation EX is followed by a subsequent sequence of time-windows including an inhalation IN. Alternatively, a breath is identified once a sequence of time-windows including an inhalation IN is following by a subsequent sequence of time-windows including an exhalation EX. The breathing analysis model may further be configured to determine a duration of an exhalation EX by counting the length of a sequence of time-windows which include an exhalation EX without including an inhalation IN. Similar considerations apply to determining the duration of an inhalation IN. The durations of an exhalation EX and / or an inhalation IN are indicative of a ventilation volume of that breath.

[0253] The breathing analysis model is configured to determine the breathing rate using a length of time between exhalations EX (in particular, between the onset of exhalations EX) and / or using a length of time between inhalations IN (in particular, between the onset of inhalations IN).

[0254] The breathing analysis model may further be configured to determine the ventilation volume by determining an envelope function enveloping the microphone signal which includes a sequence of time-windows including an exhalation EX. Analogously, the breathing analysis model may further be configured to determine the ventilation volume by determining an envelope function enveloping the microphone signal which includes a sequence of time-windows including an inhalation IN.

[0255] The breathing analysis model may further be configured to determine the ventilation volume using the spectrogram of time-windows which include an exhalation EX and / or the spectrogram of time-windows which include an inhalation IN. The relative ratio of high frequencies to low frequencies is indicative of a velocity of air travel and therefore indicative of a ventilation volume.

[0256] The chart further shows, as squares, time-points of nose-sniffing NS. The breathing analysis model may be configured to identify time-points of nose-sniffing NS as being included in a sequence of inhalations IN.

[0257] Fig. 18 shows an exemplary breathing analysis model implemented as a neural network 9. The neural network 9 comprises a plurality of layers which are connected to each other. The plurality of layers may be grouped into a one or more groups, as explained below in more detail.

[0258] The neural network 9 may be used for both regression, in which case either a sigmoid or a linear activation function may be chosen for the final layer, and / or for classification, in which a sigmoid activation function is used for binary classification, and / or a softmax function is used for multi-class classification.

[0259] Regression in this case refers to inferring continuous (e.g., floating point) values which are directly related to a value in the input. For example, a continuous output may be the number of breaths per minute. Another continuous output may relate to a tidal volume. Another continuous output may be an inferred chest expansion value or chest expansion time-series.

[0260] Classification, on the other hand, may be used to detect or infer the likelihood of particular events in the input, for example a breathing event such as an inhalation or an exhalation.

[0261] The neural network 9 may be configured to perform both classification and regression, by having two dense layers at the end, one designed for classification as described herein, and the other designed for regression as described herein.

[0262] The neural network 9 is trained using supervised learning. Supervised learning requires the use of a labelled training dataset as described herein. The labelled training dataset comprises a microphone signal and optionally a dynamic chest expansion signal, and may include labels at time-points corresponding to the start and end of an inhalation and / or an exhalation. The labelled training dataset may further include labels for other auditory events, such as caused by a person’s movement (e.g., footfalls), or due to a person’s environment (e.g., passing cars).

[0263] The labelled training dataset may further be categorized according to a subjective quality of the audio sample, as well as optionally the presence and / or amount of background noise.

[0264] The labelled training dataset may further include labels indicative of a level of effort of the user (a higher effort typically results in a difference in the inhalations and exhalation, for example a difference in the frequency spectrum and / or duration of each).

[0265] The labelled training dataset may further include labels indicative of further conditions of the user or types of breathing, for example wheezing may be indicative of a condition such as a respiratory infection or asthma. The labelled training dataset may include details of the person / user, such as gender, age, height and / or weight.

[0266] The neural network 9 may be trained, using supervised learning, to generate, from the input data included in the labelled training dataset, the labels and / or other ground truths included in the labelled training dataset, such as the chest expansion signal, the breathing rate, and / or a tidal volume.

[0267] Specifically, the neural network 9 may be trained to generate, from the microphone signal, in particular an audio snippet in the form of an amplitude time-series and / or an audio snippet in the form of a spectrogram time-series, a chest expansion signal. Thereby, the neural network 9 may, when deployed in a trained state, infer from a microphone signal, a chest expansion signal.

[0268] The (inferred) chest expansion signal can then be readily used, for example by using peak detection and / or spline interpolation, to determine a number of breaths per minute in the microphone signal.

[0269] The neural network 9 receives as an input 90 a part of the microphone signal, in particular an audio snippet. The audio snippet may be relatively short, e.g., between 0.5 seconds and 5 seconds, preferably between 0.5 and 1 second. Such a short audio snippet is suitable for detecting a breathing event, such as an inhalation or an exhalation. The audio snippet may however, be of a length sufficient to accurately determine the breathing rate from the (single) audio snippet itself, and therefore may have a length of between 3 seconds and 20 seconds, for example. Depending on the length selected, the short audio snippet may include several breaths.

[0270] The input 90 may further include location and / or movement data, for example gathered using a GPS or motion sensor such as an accelerometer or gyroscope. Thereby, the input 90 may include information related to a walking or running rhythm which, depending on the user and / or level of exercise intensity, may be used by the neural network 9 for purposes of interpolating a missing or poor quality microphone signal. This is because, for some users and / or at some exercise intensity levels, their breathing rate relates to their step rate, e.g., they may take an integer number of steps per breath.

[0271] The input 90 may include, in the form of a one dimensional vector, the amplitude of the microphone signal, in particular the amplitude of the audio snippet. The input 90 may, alternatively or additionally, include, in the form of a two dimension vector, a spectrogram of the microphone signal, in particular of the audio snippet.

[0272] As the spectrogram of an audio snippet may be represented as a two dimensional image, techniques known from image classification, such as convolutional neural networks (CNNs) may be applied as described in the following sections.

[0273] The output 91 of the neural network 9 may include a vector output 911 , which may be in the form of a one dimensional vector which may include, as parameters, probabilities or likelihoods of particular audio events being present in the audio snippet (e.g., breathing events).

[0274] The vector output 911 may also include, as part of its output, a continuous value indicative of the breathing rate.

[0275] The vector output 911 may also include, as part of its output, a (inferred) chest expansion signal which can be readily used to determine the breathing rate and / or the tidal volume as explained herein.

[0276] The vector output 911 may include a single floating point value. The vector output 911 may additionally be accompanied by a probability score indicative of a reliability of other values included in the vector output 911 . This may be determined or calculated by a suitable method which may reflect, e.g., the dropout during inference or a weighted average (optionally including a standard deviation) from prior inferences determined using prior audio snippets, which may partially overlap a current audio snippet.

[0277] The output 91 of the neural network 9 may, additionally or alternatively, include a scalar output 911. The scalar output may represent the breathing rate in the input 90.

[0278] The neural network 9 is configured to produce, using a first set of layers 92, an embedding. The output 91 includes a plurality of layers which is configured, resp., trained during supervised learning, such that the embedding is mapped to the vector output 910 and / r the scalar output 911 such that the output 91 infers the desired classes and / or parameters, e.g., whether a breathing event was detected, a type of breathing event, the breathing rate, the chest expansion signal, and / or a tidal volume.

[0279] The output 91 includes, as shown in Fig. 18, one or more densely connected layers. In particular, the output includes two densely connected layers, a first having a ReLu activation function and the second having a sigmoid activation function.

[0280] The first set of layers 92 of the neural network 9 may be considered to begin immediately downstream of the input 90 and may be considered to end at the GlobalAveragePooling step. The first set of layers 92 produce the embedding and are configured such that they have the property of similarity. The property of similarity is that when two inputs are provided which are close to each other, for example the two inputs relate to two samples of the same person wearing the same microphone or headset, the microphone signal merely being recorded in different environments, that the embedding (which may be in the form of a vector having one or more dimensions) is also close, using some notion of mathematical closeness.

[0281] The first set of layers 92 in particular comprises N (one or more) convolutional stages 93. The convolutional stage 93 is designed to extract features from the input 90, in particular relatively short time-span features which may extend across only a small number of time-slices in the input 90. The convolutional stage 93 may be implemented in a plurality of ways. Below described is one exemplary implementation of the convolution stage 93 as depicted in Fig. 18.

[0282] The convolutional stage 93 comprises, as shown in the Fig., a two dimensional convolutional layer. The two dimensional convolutional layer may be a classic two dimensional convolutional layer, or a depth wise two dimensional convolutional layer. The depth wise two dimensional convolutional layer is a type of convolution where a independent convolutional filters are applied to each input channel. The input 90 may be separated into separate input channels by way of dividing the input into two or more frequency channels and / or volume channels. A further possibility is to provide magnitude and phase information of the spectrum as two separate channels. The two dimensional convolutional layer has a linear activation function. The next layer in turn is an activation layer with a ReLu activation function (rectified linear unit) which is connected to a max pooling (2D) layer and a batch normalization layer.

[0283] Depending on the embodiment, the convolutional stage 93 may be repeated a second or third time. In other words, two or more convolutional stages 93 may be chained to each other in sequence. Each convolutional stage 93 may have different parameters, for example a different window sizes, stride lengths, etc.

[0284] Downstream from the convolutional stage 93, an attention stage 94 is arranged. The attention stage 94 is designed to process its input in sequence, in particular a sequence corresponding to time. Thereby, features related to time-spans longer than those captured by the convolutional stage 93 in the input 90 may be captured, respectively features extracted from the input 90 during the convolutional stage 90, may be determined. The attention stage 94 may be implemented in a plurality of ways, using techniques known from natural language processing such as attention and / or long shortterm memory (LSTM). Below described is one exemplary implementation of the attention stage 94 as depicted in Fig. 18.

[0285] The attention stage 94 depicted includes a reshape layer which reshapes the output from the convolutional stage 93 such that it is in a format appropriate for the subsequent multihead attention layer. Subsequent to the multi-head attention layer, the reshaped input and the output from the multi-head attention layer are multiplied in a multiply layer, which then undergoes batch normalization in a subsequent layer.

[0286] Depending on the embodiment, the attention stage 94 may be repeated a second or third time. In other words, two or more attention stages 94 may be chained to each other in sequence. Each attention stage 94 may have different parameters, for example a different number of heads, parameters, etc. In another embodiment, there is no attention stage and the model comprises only the convolutional stage and the final output (stage).

[0287] A global average pooling layer 95 follows the attention stage 94. The output of the global average pooling layer 95 is considered to provide the embedding mentioned above.

[0288] The output 91 then follows as described above, producing a vector output 910 and / or a scalar output 911. The above-described embodiments of the disclosure are exemplary and the person skilled in the art knows that at least some of the components and / or steps described in the embodiments above may be rearranged, omitted, or introduced into other embodiments without deviating from the scope of the present disclosure.

Claims

CLAIMS1. A method for determining a ventilatory threshold (VT1 , VT2) of a person (8), the method comprising: receiving (S111), in a processor (11), a recorded microphone signal from a microphone associated with a person (8) performing exercise, wherein the microphone signal was recorded during a recording period including at least two periods of different exercise intensity; determining (S112), in the processor (11), breathing rate values of the person (8) from the microphone signal, using a breathing analysis model trained using supervised learning and a labeled training dataset; and analyzing (S113), in the processor (11), the breathing rate values of the person (8) to determine a ventilatory threshold (VT1 , VT2) indicative of a separation between two ventilatory zones (Z1 , Z2, Z3) of the person (8).

2. The method according to claim 1 , the method further comprising receiving, in the processor (11), an exercise protocol indicator associated with a pre-defined exercise protocol, the pre-defined exercise protocol defining two or more levels of exercise intensity at which the person (8) is to exercise for the two or more periods, respectively; and determining, in the processor (11), the one or more ventilatory thresholds (VT 1 , VT2) using the exercise protocol indicator.

3. The method according to claim 2, the method further comprising:generating (S121), in the processor (11), a control signal for an auxiliary exercise device, the control signal defining one or more levels of resistance generated by the auxiliary exercise device according to the pre-defined exercise protocol, the control signal thereby defining the two or more levels of exercise intensity; transmitting (S122), by the processor (11), using a communication interface, the control signal to the auxiliary exercise device; and determining (S123), in the processor (11), for each of the one or more ventilatory thresholds (VT1 , VT2), an association with at least one of: a particular level of resistance or a particular level of exercise intensity at which the ventilatory threshold (VT1 , VT2) is reached.

4. The method according to one of claims 1 to 3, further comprising: receiving (S131), in the processor (11), during exercise, a performance signal from a device associated with the person (8), the further signal indicative of a measured performance level of the person (8) during exercise; and associating (S132), in the processor (11), using the further signal, the measured performance level with the one or more ventilatory thresholds (VT1 , VT2).

5. A method for determining a ventilatory zone (Z1 , Z2, Z3) of a person (8) from among a plurality of ventilatory zones (Z1 , Z2, Z3), separated by one or more predefined ventilatory thresholds (VT1 , VT2) determinable using the method according to one of claims 1 to 4, the method comprising:receiving (S141), in a processor (11), a microphone signal from a microphone associated with a person (8); determining (S142), in the processor (11), a breathing rate of the person (8) from the microphone signal, using a breathing analysis model trained using supervised learning and a labeled training dataset; and determining (S105), in the processor (11), a ventilatory zone (Z1, Z2, Z3) of the person (8), from among the plurality of ventilatory zones (Z1, Z2, Z3), using the breathing rate and the one or more pre-defined ventilatory thresholds (VT1 , VT2).

6. The method of one of claims 1 to 5, wherein the method further comprises: pre-processing, in the processor (11), the microphone signal using a digital filter, wherein the digital filter is configured to isolate, in the microphone signal, breathing-related sounds.

7. The method of one of claims 1 to 6, wherein determining the breathing rate from the microphone signal comprises: identifying (S151), in the processor (11), using the breathing analysis model, a breath by analyzing a defined time-window in the microphone signal; and determining (S152), in the processor (11), the breathing rate using a sequence of the defined time-windows, preferably an overlapping sequence of the defined time-windows, across a defined period of the microphone signal, and identifying unique breaths in the defined period.

8. The method of claim 7, the method further comprising: identifying, in the processor (11), using the breathing rate and the one or more pre-defined ventilatory thresholds (VT1, VT2), a ventilatory transition indicative of the breathing rate crossing one of the one or more pre-defined ventilatory thresholds (VT1, VT2).

9. The method according to one of claims 1 to 8, further comprising determining, in the processor (11), a physiological load on the person (8) during exercise, the physiological load determined using the breathing rate values and a period of the exercise.

10. The method according to claim 9, wherein determining the physiological load includes determining, in the processor (11), an aerobic physiological load and an anaerobic physiological load: the aerobic physiological load calculated, in the processor (11), using one or more periods during exercise in which the person is above a baseline breathing rate and below a particular ventilatory threshold (VT2) and the anaerobic physiological load calculated, in the processor (11), using one or more periods during exercise in which the person is at or above the particular ventilatory threshold (VT2).

11. An electronic system (1) comprising a processor (11) configured to perform the method according to one of claims 1 to 10.

12. The electronic system (1) according to claim 11 , wherein the electronic system (1) further comprises a microphone (2) worn on or held by the person (8), the microphone (2) configured to generate the microphone signal.

13. The electronic system (1) according to one of claims 11 or 12, wherein the electronic system (1) includes a wearable device (4) and the processor (11) is integrated into the wearable device (4).

14. The electronic system (1) according to one of claims 11 to 13, wherein the electronic system (1) includes a device (4) associated with the person (8) configured to generate a performance signal indicative of a measured performance level of the person (8) and the processor (11) is further configured to perform the method according to claim 5.

15. The electronic system (1) according to one of claims 11 to 14, wherein the electronic system (1) includes a server computer (6) and the processor (11) is integrated into the server computer (6).

16. A computer program product comprising computer program code configured to control a processor (11) such that the processor (11) performs the method according to one of claims 1 to 10.

Images