Breathing sound attribution using pure tones

Abstract

An apparatus (10) and method for attributing breathing sounds to one of two bed-users is presented, The attribution is based on the temporal coherence between breathing sounds and chest movements of the bed-user who is closer to the apparatus used for the breathing attribution. The disclosed techniques avoid the need for manual labelling of data, and for the use of multiple devices. In embodiments, it is also possible to attribute breathing sounds in which determination of temporal coherence is ambiguous, by training an acoustic similarity model with reference data for the primary and secondary bed-users.

Description

[0001] Breathing Sound Attribution Using Pure Tones

[0002] Field of Invention

[0003] The present invention relates to attribution of breathing sounds to one of two users sharing a bed. More particularly, the present invention relates to an apparatus for attributing breathing sounds to bed-users positioned at different distances from the apparatus.

[0004] Technical Background

[0005] Smartphone applications that record and analyse breathing sounds have become increasingly popular among consumers and healthcare professionals as tools for gaining insights into sleep quality and potential sleep-disordered breathing. These applications typically utilize a device's microphone to measure audible sound frequencies, with a focus on increased sound intensity in specific frequency ranges as a characteristic indicator of snoring. In addition, breathing sounds, that may be very quiet, are useful for predicting sleep stages such as rapid eye movement (REM).

[0006] A significant challenge faced by sleep audio analysis applications is the accurate attribution of breathing sounds when two people share a bed, which is difficult to achieve using audio analysis alone. The variability in breathing sounds from a single user throughout the night, influenced by factors such as sleeping position, distance from the microphone, and anatomical locations of any airway resistance, further complicates the attribution process.

[0007] One approach to addressing this problem involves the use of two paired devices positioned on either side of the bed, sharing acoustic signal information to attribute breathing sounds based on relative loudness or sound profile. However, this solution requires additional hardware in the form of a second recording device, creating inconvenience for users. It also requires complex data synchronisation processes between the devices. Moreover, attribution based solely on relative loudness may lead to false attributions due to factors such as body position, room acoustics, and microphone variations. For example, a user lying on their side and facing the direction of their partner’s device may create a louder signal at their partner’s device than at their own.International patent application WO 2023 / 247436 discloses a potential solution using a single device, wherein the user manually labels numerous snoring recordings made during the night. This labelled data is then used to train a machine learning model for retroactive estimation of snore origins. While this method eliminates the need for multiple devices, it introduces a time-consuming and potentially error-prone manual labelling process, and the resulting attribution model may still make errors due to similarities between partners' breathing sounds and other complexities.

[0008] The present invention is developed in this context, providing a breathing sound attribution technique that overcomes one or more of the above problems.

[0009] Summary of Invention

[0010] According to an aspect of the present invention, there is provided an apparatus for attributing breathing sounds to one of two bed-users positioned at different distances from the apparatus, wherein the apparatus comprises: a speaker arranged to output a pure tone; a microphone for detecting an audio signal and for detecting a reflection of the pure tone; and a controller arranged to: estimate chest movements of a primary beduser, nearer to the apparatus than a secondary bed-user, from the reflection of the pure tone; detect breathing sounds by acoustic analysis of the audio signal; determine a degree of temporal coherence between the detected breathing sounds and the estimated chest movements; attribute the origin of the breathing sounds to either the primary or secondary bed-user in dependence on the degree of temporal coherence; determine one or more states of the primary and / or secondary users from the classified breathing sounds; and output data associated with the determined one or more states.

[0011] In embodiments, the one or more state comprise one or more of snoring, respiratory rate, breathing disturbance, sleep stage estimate, and awakeness.

[0012] In embodiments, the controller is arranged to further perform: labelling of breathing sounds with confidence scores; and construction of reference data sets for each user comprising a subset of the breathing sounds labelled with high confidence scores for the respective user; wherein the reference data sets are for training of an acoustic similarity model for determining whether a breathing sound is similar to a breathing sound in the reference data set for each user.In embodiments, the confidence scores are determined based on a signal to noise ratio of the reflection of the pure tone.

[0013] In embodiments, the controller is arranged to use the acoustic similarity model to reattribute the breathing sounds, or to attribute further breathing sounds, to the primary or secondary bed-user based on acoustic similarity to breathing sounds in the reference data set for that user.

[0014] In embodiments, the apparatus comprises a motion-estimation module for estimating chest movements of a user based on the reflected pure tone.

[0015] In embodiments, the motion-estimation module is arranged to estimate motion of the chest of the user based on a Doppler shift of the frequency of the reflected pure tone, relative to a pilot frequency of the pure tone.

[0016] In embodiments, the microphone comprises a first microphone arranged to receive the audio signal, and a second microphone arranged to receive the reflected pure tone.

[0017] In embodiments, the apparatus comprises the analysis engine, wherein the analysis engine is a sleep analysis engine arranged to identify one or more sleep patterns for one or more users of the plurality of users based on the audio data attributed to the one or more users.

[0018] According to a second aspect of the present invention, there is provided a method of attributing breathing sounds to one of two bed-users positioned at different distances from an apparatus, comprising: outputting a pure tone; detecting an audio signal and for detecting a reflection of the pure tone; and estimating chest movements of a primary bed-user, nearer to the apparatus than a secondary bed-user, from the reflection of the pure tone; detecting breathing sounds by acoustic analysis of the audio signal; determining a degree of temporal coherence between the detected breathing sounds and the estimated chest movements; attributing the origin of the breathing sounds to either the primary or secondary bed-user in dependence on the degree of temporal coherence; determining one or more states of the primary and / or secondary users from the classified breathing sounds; and outputting data associated with the determined one or more states.According to a third aspect of the present invention, there is provided a computer program which, when executed by one or more processors, is arranged to perform the method of the second aspect.

[0019] In this way, breathing sounds are automatically attributed to one of two bed-users, based on their temporal coherence with the chest movement of the bed-user who is closer to the apparatus used for the breathing attribution. The disclosed techniques avoid the need for manual labelling of data, and for the use of multiple devices. In embodiments, it is also possible to attribute breathing sounds in which determination of temporal coherence is ambiguous, by training an acoustic similarity model with reference data for the primary and secondary bed-users.

[0020] Brief Description of Drawings

[0021] Embodiments of the present invention will be described by way of example only, with reference to the accompanying drawings, of which:

[0022] Figure 1 shows the arrangement of an apparatus according to embodiments of the present invention;

[0023] Figure 2 shows an example of high temporal coherence between breathing sounds and chest movements;

[0024] Figure 3 shows an example of low temporal coherence between breathing sounds and chest movements; and

[0025] Figure 4 shows a breathing sound attribution method according to embodiments of the present invention.

[0026] Detailed Description

[0027] Embodiments of the present invention are described herein in the context of a smartphone, hosting an apparatus for detecting breathing sounds and attributing the breathing sounds to one of two bed-users. Figure 1 illustrates a configuration of the apparatus 10 according to embodiments of the present invention. The apparatus 10 is positioned to one side of a bed in which two users are sleeping. In the present disclosure, the smartphone is referenced as being arranged on the side of the bed closest to a first or ‘primary’ bed user 11, who may be the main user of the smartphone. In this context, the disclosed technique provides a solution to the challenge of attributing the breathingsounds 18 to the correct bed-user, and avoids the need for use of two devices or a manual labelling process.

[0028] The apparatus 10 comprises a speaker 13, a microphone 14, and a controller 15. The speaker 13 is controlled to output a pure tone signal 16. The microphone 14 detects reflection 17 of the pure tone signal, and also detects audio signals 18 including breathing sounds. The controller 15 attributes the breathing signals 18 to the first user 11 or a second user 12 based on the determination of a synchronisation between the breathing sounds 18 and the reflection 17 of the pure tone.

[0029] The pure tone 16 is a high-frequency signal emitted towards the bed, having a frequency referred to herein as ‘suprasonic’, which describes frequencies close to the 20 kHz ultrasonic threshold, e.g. from 18 kHz upwards, which are inaudible to most humans and are within the acoustic bandwidth of common devices such as smartphones. When output as a pure tone, such frequencies are not disturbing to sleep and do not interfere with recording of audible breathing sounds at lower frequencies.

[0030] Breathing involves chest movements that draw air into and expel air from the lungs, with breathing sounds arising due to tissue vibrations and airway obstructions. Breathing sounds vary based on whether air is being inhaled or exhaled, and can also vary in intensity at different points in the breath. For example, sound intensity usually peaks in the latter part of an inhalation and then diminishes when the chest movement inflects into an exhalation, inverting the direction of movement of the chest with respect to the microphone. Exhalations are typically quieter than inhalations, and may be associated with other distinctive sounds. The intensity of breathing sounds also tends to increase when larger quantities of air are drawn in, which will correlate with larger chest movements.

[0031] As each bed-user breathes, their chest moves closer to, and farther from, the microphone 14 of the apparatus 10. A portion 17 of the emitted pure tone 16 is reflected back from the primary bed-user’s chest to the microphone 14, and the distance travelled by the reflection will vary based on the distance of the chest wall from the microphone 14. The subtle variations in chest distance over a breathing cycle lead to subtle magnitude and phase differences at, and around, thefrequency of the pure tone 16 which, after sampling by the microphone 14, can be analysed over time and used to construct a signal whichis correlated to the user’s chest movement. The use of a suprasonic pure tone 16 thus advantageously enables its reflection 17 to serve as a proxy for chest movements.

[0032] Suprasonic signals are more readily scattered and absorbed by objects, such as the body of the primary user 11 , compared with lower frequency acoustic signals. As a result, any reflection from the secondary bed-user 12 is significantly attenuated in comparison with the reflection from the primary bed-user 11. In practice, the portion of the received signal which has been reflected by the secondary bed-user’s chest is negligible.

[0033] The general principle of operation of embodiments of the present invention involves using the apparatus 10 to analyse the reflection of the suprasonic pure tone signal 17 in conjunction with the audio signal in order to determine whether breathing sounds 18 are likely to have originated from the determined chest movements. Generally, a greater degree of synchronisation or temporal coherence between chest movements and breathing sounds 18 will suggest there is a higher likelihood that the breathing sounds 18 originate from the user 11 closer to the apparatus 10.

[0034] Conversely, if breathing sounds 18 are not coherent with the determined chest movements, there is a higher probability that the sounds originate from the secondary bed-user 12.

[0035] Attributed breathing sounds 18 can be fed to a sleep analysis engine or application for making one or more inferences or decisions from the data, such as determination of one more states of the user. For example, an interface showing snoring recordings, sleep stages, and metrics or statistics for each user can be displayed. The states may include at least one of snoring, breathing disturbances, measured respiratory rate, a sleep stage estimate, and a measure of awakeness.

[0036] The apparatus 10 may be a standalone device. The operation of the controller 15 is alternatively implemented as part of a software application such as that hosted by a smartphone or tablet device, which can access the device’s speaker and microphone. For example, the device's built-in speaker can be used to emit the suprasonic pure tone 16, while the device's microphone can capture both the breathing sounds 18 and the reflected tone signal 17.The apparatus 10 may be arranged to store data characterising the attributed breathing sounds for one or more bed-users, enabling analysis of user(s)’ sleep patterns. The data may include various types of information, including one or more of the following: raw audio recordings; timestamps for one or more audio events such as sharp intakes of breath; and duration, intensity, volume levels and frequency characteristics of breathing patterns.

[0037] The degree of temporal coherence between breathing sounds 18 and the chest movements may be used to determine whether breathing sounds 18 are more or less likely to originate from the primary bed-user 11 whose chest movements are derived from the reflected pure tone 17. The apparatus 10 may employ various signal processing and pattern recognition techniques to determine the temporal coherence. For example, intensity of breathing sounds and the magnitude of chest movements can be compared to determine a level of consistency of breathing patterns over a period of time. The period of time is advantageously typically at least 15 seconds, so as to include more than one breath cycle. The apparatus 10 determines synchronisation between the audio signal and changes in chest position as derived from the reflected pure tone and outputs a degree of temporal coherence.

[0038] Figure 2 illustrates an example of high temporal coherence observed between the reflection 17 of the pure tone and breathing sounds 18. The upper trace 21 shows variation in the magnitude of the reflection 17 of the pure tone signal with time. Seven breath cycles are observed. The variation in magnitude corresponds to chest movements, as described above. The lower trace 22 shows breathing sounds 18 in the audio signal, as represented by their acoustic power. Again, seven audible breathing sound cycles are shown. In the example of Figure 2, the spacing between the occurrences of breathing sounds and the spacing between the chest movements is very similar, and the breathing sounds are temporally synchronised with the latter part of an inhaling motion, represented by magnitude of the tone reflection. Such temporal coherence between the upper 21 and lower 22 traces infers a high likelihood of that the captured breathing sounds 18 originate from the first user 11 closer to the apparatus 10.

[0039] In contrast, Figure 3 shows an example of low temporal coherence observed between the reflection 17 of the pure tone and breathing sounds 18. Here, there are eight chest movement cycles, but only six audible breathing sound cycles. The spacing betweenbreathing sounds 18 and the peaks of the chest movements are not the same such that the breathing sounds 18 and chest movements are not synchronised. Such poor temporal coherence between the upper 31 and lower traces 33 infers that the captured breathing sounds 18 originate from the second user 12 further from the apparatus 10.

[0040] Temporal coherence can be derived based on the numbers of breath cycles and chest movement cycles, over a period of time, being the same. However, alternative measures of temporal coherence can be derived using any appropriate pattern processing algorithm.

[0041] In embodiments, a binary determination as to whether breathing sounds 18 are more likely to be associated with the primary bed-user 11 or the secondary bed-user 12 can be made, based on whether or not the breathing sounds 18 are coherent with the reflected pure tone 17. In some scenarios, the confidence in such a determination may be low, particularly if there is a low signal to noise ratio in the reflected pure tone 17. This can arise if there is additional noise or sub-optimal positioning of the apparatus 10. In such cases, the confidence in the attribution of breathing sounds 18 can be increased by the introduction of further analysis based on the acoustic features of the breathing sounds 18.

[0042] In such embodiments, the attribution stage is informed using subsets of breathing sounds 18 for which the degree of coherence is demonstrated clearly, indicative of likely user origin. Breathing sounds for which the level of coherence is not demonstrated clearly, indicating ambiguous origin, are excluded. A clear demonstration of the degree of coherence can be associated with a high signal to noise ratio in the reflected pure tone 17. The subsets of breathing sounds are used as reference data sets of each bed-user’s breathing sounds, for configuration of any appropriate acoustic similarity model, such as a spectral similarity model, to be implemented in the attribution stage. Breathing sounds are attributed to either user based on acoustic similarity to breathing sounds in the reference data sets.

[0043] The acoustic similarity attribution can be retrospectively applied to previously attributed breathing sounds, which may result in a change in attribution of some of the breathing sounds. Alternatively, the acoustic similarity attribution can be applied to new breathing sounds captured in future contexts.Figure 4 illustrates a breathing sound attribution method according to embodiments of the present invention. The method may be a computer-executable method, executed by the controller 15 of the apparatus 10 shown in Figure 1.

[0044] The method begins with the receipt of an audio signal by the microphone 14 of the apparatus 10, and the detection of breathing sounds 18 from the received audio signal in step S41. Any suitable acoustic analysis algorithm may be used to identify breathing sounds 18 from the audio signals, including spectral analysis and comparison with known frequencies of breathing sounds.

[0045] At the same time as step S41 , the microphone 14 receives a reflection 17 of a pure tone 16 emitted by the speaker 13 of the apparatus 10, and determines chest movements of the primary bed-user 11 in step S42. The breathing sounds and chest movements are measured with respect to a common timeframe, so that it is possible to determine in step S43 whether there is temporal coherence between the breathing sounds and the chest movements. Temporal coherence is determined in the manner described above, by assessing the degree of synchronisation between the chest movements and the breathing sounds 18. Any suitable algorithm may be used in order to determine whether, features of a chest movement signal and breathing sounds are synchronised.

[0046] In step S46, a degree of confidence in the temporal coherence identified in step S43 is determined. Generally, step S46 can be considered as having three outcomes:

[0047] (a) a high degree of temporal coherence is clearly demonstrated, indicating high likelihood that the breathing sounds 18 originate from the primary bed-user 11;

[0048] (b) a low degree of temporal coherence is clearly demonstrated, indicating high likelihood that the breathing sounds 18 originate from the secondary bed-user 12;

[0049] (c) the degree of temporal coherence is not clearly demonstrated, leading to ambiguity in the determined origin of the breathing sounds 18.

[0050] For cases (a) and (b), clear determination of the degree of temporal coherence arises from strong breathing signals, with a low degree of noise. Case (c) arises where noise levels are higher, so that, irrespective of the degree of coherence, it is more difficult to determine with confidence whether the breathing sounds 18 originate from the primary or 11 the secondary bed-user 12.In some embodiments, all breathing sounds 18 are attributed in step S44, based on a comparison of a degree of temporal coherence with a threshold, such that even for case (c), an attribution to either the primary 11 or secondary bed-user 12 is performed. Such embodiments conclude with the determination of the state of the primary and / or second bed-user 11, 12 in step S45, with the results output to a user in the manner described above. Such embodiments are illustrated via the dotted lines from step 43.

[0051] In the present embodiment, however, the method continues to train and use an acoustic similarity model to produce more confident attributions of the breathing sounds 18, and in step S46, a determination of the confidence of the attribution made possible by the determination of temporal coherence in step S43 is performed. Where confidence is low, associated with case (c), breathing sounds 18 are not attributed and unattributed breathing sounds 51 are stored for attribution.

[0052] Reference breathing sounds 47 are constructed for the primary and secondary bed-users 11, 12, containing breathing sounds 18 identified from cases (a) and (b) in step S46, namely breathing sounds attributed to the primary 11 or secondary 12 bed-users with a high degree of confidence. In step S48, the reference breathing sounds 47 are used to train an acoustic similarity model. In some embodiments, step S48 comprises cleaning or reattributing reference breathing sounds which are not consistent with the majority of breathing sounds in the reference group.

[0053] In step S49, the acoustic similarity model is used to attribute new breathing sounds, detected in a new context in step S50, such as on a different night from the breathing sounds 19 captured in step S41. In step S49, the acoustic similarity model is used to determine whether newly detected breathing sounds identified in step S50 are acoustically similar to those in reference data sets 47 for either the primary 11 or secondary 12 data sets.

[0054] In alternative embodiments, the unattributed sounds 51 identified in step S46 are input to the acoustic similarity model for attribution in step S49 instead of, or in addition to newly detected breathing sounds from step S50. This enables resolution of the ambiguity in the attribution of the breathing sounds 18 in step S46.Breathing sounds which are determined to be acoustically similar to those contained in the reference breathing sounds 47 are attributed to the bed-user associated with the acoustically similar reference breathing sounds, and this is output as an attribution result in step S49, which informs a state determination and output step S52.

[0055] Features of any of the embodiments outlined above may be combined to create additional embodiments, falling within the scope of the claims. It will also be appreciated that a number of modifications to the described embodiments are possible.

[0056] For example, the microphone may be a single component capable of capturing both audible frequencies and the suprasonic pure tone signal, or separate microphones may be used for audio detection and suprasonic pure tone signal detection. Separate microphones may lead to improved signal quality, as each microphone can be optimized for its specific frequency range, potentially resulting in clearer audio recordings and higher accuracy. Further, reduced interference may be possible. By separating the audio and suprasonic pure tone signal detection, it is possible to minimize potential cross-talk or interference between the two signal types. Flexibility in microphone placement is such that an audio microphone can be placed closer to the user's head, while the suprasonic signal microphone can be directed towards the user's chest.

[0057] Although reference is made herein to breathing sounds, inferences may also be drawn from a lack of breathing sounds, particularly with regard to irregular patterns associated with sleep-disordered breathing. For example, obstructive sleep apnoea events may be characterised by chest movements with an absence of breathing sounds, whilst central sleep apnoea events may be characterised by reduced chest movements.

[0058] In embodiments, data may be added to the reference data sets for each user from test processes performed in single bed-user environments, in which it can be determined with confidence that a detected breathing sound is associated with the bed-user.

[0059] In embodiments, temporal coherence may be estimated using a neural network. The process of training the neural network may involve exposing the neural network to a diverse set of audio samples from various users. Some of these will be from single bedusers in which chest movements and breathing sounds are expected to be synchronised, but it is possible to add breathing sounds and chest movements that are known not tocorrespond to each other (for example, signals from different users, or signals received at different times). The neural network is trained to be able to identify temporally coherent chest movements and breathing signals, and to distinguish these from signals that are not coherent. These samples include different types of breathing sounds, such as normal breathing, snoring, and other sleep-related respiratory events, and with varying degrees of background noise.

[0060] In some embodiments, the frequency used for the pure tone signal may be adjustable. This adaptability can allow the system to optimize performance based on factors such as room acoustics, user physiology, or environmental conditions.

[0061] In some embodiments, the apparatus includes a dedicated motion-estimation module for estimating breathing motion from the reflected suprasonic pure tone, rather than implementing motion-estimation functionality in the controller. The motion-estimation module may utilize various algorithms and signal processing techniques to accurately estimate the magnitude and timing of chest movements associated with breathing.

[0062] In some embodiments, a method employed by the motion-estimation module for estimating motion is based on the Doppler shift of the reflected suprasonic pure tone. As the user's chest moves during breathing, the frequency of the reflected pure tone changes slightly due to the Doppler effect. By measuring the frequency shift of the reflected signal relative to the original emitted signal (pilot frequency), the motionestimation module can calculate the velocity and displacement of the chest wall. This technique is also useful for determining the sleep stage of the user and larger body movements associated with awakeness.

[0063] The principles of motion estimation, based on analysis of a pure tone, are not reproduced in detail herein, in the interests of conciseness, and are set out in detail in the inventor’s prior patent applications such as international patent application WO 2022 / 185025. In embodiments of the present invention, any motion-estimation technique, with which a breathing pattern can be inferred from a signal representing chest movements, is suitable for use.

Claims

1. Claims1. An apparatus for attributing breathing sounds to one of two bed-users positioned at different distances from the apparatus, wherein the apparatus comprises:a speaker arranged to output a pure tone;a microphone for detecting an audio signal and for detecting a reflection of the pure tone; anda controller arranged to:estimate cyclical chest movements of a primary bed-user, nearer to the apparatus than a secondary bed-user, from the reflection of the pure tone;detect cyclical breathing sounds by acoustic analysis of the audio signal; determine a degree of temporal coherence between the detected cyclical breathing sounds and the estimated cyclical chest movements, based on a comparison of the number of detected breathing sound cycles and estimated chest movement cycles occurring over a predetermined time period;attribute the origin of the cyclical breathing sounds to either the primary or secondary bed-user in dependence on the degree of temporal coherence; determine one or more states of the primary and / or secondary users from the attributed breathing sounds; andoutput data associated with the determined one or more states.

2. An apparatus according to claim 1 , wherein the one or more states comprise one or more of snoring, respiratory rate, breathing disturbance, sleep stage estimate, and awakeness.

3. An apparatus according to claim 1 or claim 2, wherein the controller is arranged to further perform:labelling of breathing sounds with confidence scores; andconstruction of reference data sets for each user comprising a subset of the breathing sounds labelled with high confidence scores for the respective user;wherein the reference data sets are for training of an acoustic similarity model for determining whether a breathing sound is similar to a breathing sound in the reference data set for each user.

4. An apparatus according to claim 3, wherein the confidence scores are determined based on a signal to noise ratio of the reflection of the pure tone.

5. An apparatus according to claim 3 or claim 4, wherein the controller is arranged to use the acoustic similarity model to reattribute the breathing sounds, or to attribute further breathing sounds, to the primary or secondary bed-user based on acoustic similarity to breathing sounds in the reference data set for that user.

6. An apparatus according to any one of the preceding claims, comprising a motionestimation module for estimating chest movements of a user based on the reflected pure tone.

7. An apparatus according to claim 6, wherein the motion-estimation module is arranged to estimate motion of the chest of the user based on a Doppler shift of the frequency of the reflected pure tone, relative to a pilot frequency of the pure tone.

8. An apparatus according to any one of the preceding claims, wherein the microphone comprises a first microphone component arranged to receive the audio signal, and a second microphone component arranged to receive the reflected pure tone.

9. An apparatus according to any one of the preceding claims, comprising the analysis engine, wherein the analysis engine is a sleep analysis engine arranged to identify one or more sleep patterns for one or more users of the plurality of users based on the audio data attributed to the one or more users.

10. A method of attributing breathing sounds to one of two bed-users positioned at different distances from an apparatus, comprising:outputting a pure tone;detecting an audio signal and detecting a reflection of the pure tone; and estimating cyclical chest movements of a primary bed-user, nearer to the apparatus than a secondary bed-user, from the reflection of the pure tone;detecting cyclical breathing sounds by acoustic analysis of the audio signal; determining a degree of temporal coherence between the detected cyclical breathing sounds and the estimated cyclical chest movements based on a comparison- 15 -of the number of detected breathing sound cycles and estimated chest movement cycles occurring over a predetermined time period;attributing the origin of the breathing sounds to either the primary or secondary bed-user in dependence on the degree of temporal coherence;determining one or more states of the primary and / or secondary users from the attributed breathing sounds; andoutputting data associated with the determined one or more states.

11. A computer program which, when executed by one or more processors, is arranged to perform the method of claim 10.

Images