Audiometer

Abstract

A method of audiometry comprises outputting a dynamic audio stimulus having a plurality of frequencies to a subject S100, filtering the stimulus to a selected subset of the frequencies S102, monitorin

Description

FIELD OF THE INVENTION The invention relates to a method and apparatus for controlling and monitoring an acoustic stimulus and detecting a response in the field of audiometry. BACKGROUND Audiometry is the assessment of the human biological hearing mechanism. More specifically, by audiometry involves the assessment of a subject’s ability to hear particular frequencies at particular intensities, or volumes. In general, people’s hearing is more sensitive at some frequencies and less sensitive at others. In some cases, this means that the hearing is sufficiently impaired in some regions of the frequency spectrum to require the provision of hearing aids. Determining the level of hearing a person has at different frequencies is required to understand the auditory capabilities and limitations of the subject. It is known in the art how to assess hearing with trained, professional audiologists carrying out audiometric hearing tests, presenting stimuli to a subject and recording whether the subject indicates they have heard the stimulus or not. This method requires the subject to be a willing, engaged participant of the hearing test, co-operating with the audiologist to assess their level of hearing. However, some subjects are unresponsive to traditional audiometric stimuli, such as monotonic tones or beeps or narrowband stimuli such as warble-tones or narrow-band noises. These uncomplicated stimuli are traditionally used because they contain one or limited frequency components, which helps isolate and determine which frequencies the subject can hear and because they are static, their intensity doesn’t change over time — allowing a determination of their relative level of hearing. Use of a more complex broadband stimulus makes it difficult to ascertain which frequency component the subject is responding to. And if the stimulus is dynamic, i.e. changes with time, it is difficult to ascertain the hearing level of a subject. In certain special cases, some subjects, such as children with learning difficulties, are unable to participate and provide affirmation of hearing in traditional audiometric approaches. This makes traditional approaches of audiometry much more difficult, as the audiologist must assess whether it appears the subject has reacted to the stimulus without a direct confirmation from the subject themselves. There is therefore a need for a method of audiology which is compatible with more complex stimuli and does not require the active participation of the subject. SUMMARY OF INVENTION According to a first aspect, there is provided a computer-implemented method of audiometry, the method comprising: outputting a dynamic audio stimulus to a subject, the dynamic audio stimulus comprising a plurality of different frequencies; applying a filter to the dynamic audio stimulus as it is output, to restrict the dynamic audio stimulus to a selected subset of the frequencies; monitoring input data and determining a response of the subject at a response time based on the input data; and recording an amplitude of one or more frequencies within the subset of frequencies at a measurement time, the measurement time determined based on the response time. The method presented herein allows for the assessment of the subject’s hearing across a full frequency spectrum, using a dynamic audio stimulus with multiple frequency components at different amplitudes. In general, the amplitudes of the different frequency components in the stimulus vary as a function of time, hence “dynamic”. The different frequency components may span the audible hearing range, or may be in a subset of it, such as 0.1 to 10 kilohertz. The product of an audiometric study is traditionally called an “audiogram” and represents the hearing threshold, i.e. the amplitude at which each frequency begins to be heard by the subject. The method is particularly effective in assessing the hearing of subjects who are not responsive to traditional methods of audiometry. These subjects may be responsive to some sounds, such as a particular piece of music or the theme tune of a favourite television show. Traditional audiometry cannot use such a stimulus, due to the multiple frequency components and their shifting amplitudes over time. This makes it very difficult for an audiologist to discern what part of the stimulus the subject has responded to, if any. The method includes a filtering step which, amongst other purposes, may help simplify the use of a complex dynamic stimulus. The dynamic audio stimulus can be restricted to a subset of frequencies present in the original, before being outputted to the subject. The selected frequencies may be changeable in real time. Selecting different frequency ranges may allow for testing a subject’s response to particular parts of a dynamic audio stimulus, in order to accurately assess exactly which frequencies elicit a response. They may also be filtered (e.g. compressed) to limit the intensity variations of the dynamic signal. Filtering the dynamic audio stimulus in this way may preserve the original character of the stimulus, allowing it to still elicit a response from patients who are not responsive to other stimuli. The response of the subject is determined by monitoring input data. The input data is usable to measure a subject’s response to the dynamic audio stimulus. The input data may comprises data usable to measure a subject’s physical response to the dynamic audio stimulus. The input data preferably comprises data indicative and / or predictive of the subject’s response to the stimulus. In particular the input data comprises data usable to measure the subject’s response to the studio stimulus. The input data may be referred to as “subject data”. The input data may comprise a direct input from the subject, i.e. a manual input such as a button press or vocal confirmation. The input data may alternatively or additionally comprise photo or video data of the subject, or physiological measurements of the subject such as heart rate or brain activity. The time at which a response is detected based on the input data is the response time. Amplitudes of the frequencies in the filtered dynamic audio stimulus outputted are continuously monitored and recorded at a measurement time, corresponding to the response time. As the stimulus is dynamic, it may be important to record the amplitudes of the various frequency components that the subject was hearing when they responded. Preferably the measurement time is at or before the response time. Where the measurement time is before the response time it may be calculated by subtracting a response lag time from the response time. The response lag time my be selected based on one or more subject parameters, for example age. The dynamic audio stimulus may be stopped once the response has been determined. The stopping may occur immediately upon determination of a response, or with a delay. The dynamic audio stimulus may be output using a speaker. Such a speaker may be calibrated to the room that the method is carried out in, to compensate for differences in acoustic response in different spaces. Alternatively or additionally, the dynamic audio stimulus may be output using headphones, which may be worn by the subject. Based on the recorded amplitudes of the frequencies in the selected subset of frequencies, a hearing threshold of the subject may be determined. A hearing threshold may be determined for a particular frequency or range of frequencies. The hearing threshold may be a predicted threshold, based on the recorded amplitude. The hearing threshold may be defined as the sound level below which a person’s ear is unable to detect any sound. Alternatively, the hearing threshold may be defined as lowest amplitude at which a particular frequency is perceived 50% of the time by the subject. The hearing threshold may be simply taken as the recorded amplitude for each frequency, may be determined based on the recorded amplitude for each frequency, or may be determined based on multiple recorded amplitudes corresponding to multiple responses. This may comprise taking the lowest amplitude from a multitude of amplitudes corresponding to a multitude of responses, for a particular frequency. Preferably, the method further comprises determining a plurality of responses at different measurement times during output of the filtered dynamic audio stimulus, for each determined response, recording an amplitude of one or more frequencies within the subset of frequencies at a measurement time, the measurement time at or before the response time and determining a hearing threshold for the subset of frequencies based on the amplitudes recorded across the plurality of responses. Optionally, the method further comprises monitoring input data while applying a plurality of different filters to the dynamic audio stimulus, each filter restricting the dynamic audio stimulus to a different subset of frequencies; determining a plurality of responses at different measurement times while applying each filter; recording an amplitude of one or more frequencies within the subset of frequencies for each of the plurality of responses determined for each of the plurality of filters; and determining a hearing threshold for each subset of frequencies based on the recorded amplitudes. Alternatively or in addition, a hearing threshold range may be determined. This hearing threshold range may comprise a lower bound and / or an upper bound, corresponding to an estimated minimum and maximum amplitude for the hearing threshold. The hearing threshold may be referred to as a “behavioural hearing threshold”, indicating that the hearing threshold refers to the threshold at which a behavioural response is detected. Changing the filtering of the dynamic audio stimulus whilst carrying out the method allows for identifying which particular part of a stimulus caused a reaction by the subject. Accordingly, the method may further comprise receiving a user input while outputting the dynamic audio stimulus and selecting the subset of the frequencies based on the user input. The subset of frequencies may be a defined frequency range, such as 100 Hz (Hertz) to 1000 Hz, or a defined range in nonlinear units, such as an octave. Alternatively, the subset of frequencies may comprise a set of frequency bands, of fixed or differing sizes. The size, e.g. one octave, and number of bands, e.g. seven bands, may be a fixed parameter, or it may be configurable. Preferably the filter comprises a variable filter that may be adjusted during assessment of a subject, i.e. the filter may be varied dynamically to adjust the frequencies selected. It is possible that the subject may not hear the dynamic audio stimulus as filtered. Therefore, the method may further comprise increasing the amplitude of the dynamic audio stimulus until the response of the subject is determined. This amplification of the dynamic audio stimulus may be applied equally across all frequencies, or biased towards some frequencies over others. The response of the subject be delayed from when they heard the part of the dynamic audio stimulus to which they are responding. Further, they may be responding to a part of the dynamic audio stimulus which only lasted for a short time, such as a momentary increase in amplitude, e.g. a cymbal crash. In order to better capture such a case, the method may further comprise measuring and recording the amplitude of each frequency within the subset of frequencies at a plurality of measurement times during a response measurement window, the response measurement window determined based on the response time. Making multiple measurements better characterises the dynamic audio stimulus in the period that elicited the response and allows for better analysis of why the subject responded. The measurement window may correspond to the human auditory window of loudness estimation, being 70 milliseconds long, or being between 50 to 100 milliseconds long, or being up to 200 milliseconds long, or up to 500 milliseconds long. The response window may end at the response time, and start some period before the response time, e.g. any of the aforementioned response window lengths. These multiple measurements in a measurement window may be analysed by recording the maximum amplitude of each frequency, thereby recording the loudest possible dynamic audio stimulus in the window. In this way, it can be ensured the hearing threshold is not underestimated. Additionally or alternatively, the multiple measurements may be analysed by recording the average amplitude measured for each frequency. In this way, a measurement representative of the whole measurement window can be created. The average amplitude in the window may be calculated as an exponential average with a time constant equal to the length of the window. The average amplitudes may be displayed as a histogram, where the width of the histogram indicates the uncertainty in the measurement. Further, in order to better determine which frequencies of the dynamic audio stimulus the subject is responding to, the method may further comprise measuring and recording the amplitude of each frequency within the subset of frequencies at a plurality of non-response measurement times during a non-response measurement window, the non-response measurement window determined based on the response time; and determining a response frequency based on the amplitudes recorded during the response measurement window and the nonresponse measurement window. The non-response window may be before or after the response window. The nonresponse window comprises a passage of time in which the subject does not respond to the dynamic audio stimulus. The non-response window may be a window separated from a response time by a pre-determined threshold. In some cases, the non-response window and the response window may overlap, but in the simplest case they are sequential. The non-response window does not contain the response time. There may be a gap between the non-response window and the response window, or one may follow immediately after the other. The hearing threshold may be determined based on the amplitudes recorded during the response measurement window (response amplitudes), and the amplitudes recorded during the non-response measurement window (nonresponse amplitudes). In particular a hearing threshold for a particular frequency or frequency band may be determined based on a maximum amplitude of the frequency or frequency band within the response window and the maximum amplitude of the frequency or frequency band recorded within the non-response measurement window. Alternatively or in addition, a hearing threshold range may be determined based on the response amplitudes and the non-response amplitudes. The response amplitudes may be used to determine an upper bound on the hearing threshold, and the non-response amplitudes may be used to determine a lower bound on the hearing threshold. Alternatively, multiple non-response windows may be used. Each non-response window may be of a different length, may be sequential or not with respect to other non-response windows, and may be before or after the measurement window. One example implementation uses a non-response window immediately before the measurement window, and a non-response window immediately after the measurement window.. The non-response window may represent a portion of the dynamic audio stimulus that the subject did not respond to. Accordingly, measurements of the amplitudes of the frequencies during the non-response window may represent amplitudes at which the subject could not hear the corresponding frequency. Accordingly, by analysis of the recorded amplitudes in the non-response window and the response window, it may be determined which frequencies the subject responded to, and / or which frequencies it is likely the subject responded to. If the amplitudes recorded for a particular frequency or frequency band are similar in the response window and the non-response window ( / .e. within some threshold of each other), it may be determined that the subject did not respond to this particular frequency or frequency band. Similarly, if the amplitudes recorded for a particular frequency or frequency band are significantly higher in the response window than the non-response window, it may be determined that the subject did respond to this particular frequency or frequency band. In this way, the hearing threshold may only be determined for frequencies that the subject responded to, without determining the hearing threshold for frequencies they did not respond to. This may improve the accuracy of the determined hearing threshold. Further, this analysis may be via analysis of histograms of the amplitude of a frequency during the non-response window and the response window. Such a “non-response histogram” and “response histogram” may be analysed with statistical methods, such as the calculation of distance measures between the two histograms, such as the Hellinger distance, Manhattan distance, Euclidean distance, Chybyshev distance, Fractional or Minkowski distance, or histogram intersection, Kolmogorov-Smirnov test, chi-squared tests, bi-histogram calculation or two-sample t-tests. Alternatively or in addition, the statistical methods may comprise calculation of a receiver operating curve and the calculation of the area under the curve and / or the D-prime statistic. A larger statistical difference between the non-response histogram and the response histogram for a particular frequency may indicate that this frequency was more relevant to the response of the subject. A smaller statistical difference between the non-response histogram and the response histogram may indicate that this frequency was less relevant to the response of the subject. Further, statistical analysis may be used to make corrections to the response time. The response time as determined from monitoring input data may have inaccuracies due to varying speed of response of the subject. The response time may be corrected by modulating the start / end times of the non-response window and response window and calculating corresponding statistics of the response and non-response histograms. In some examples, the response time may be corrected by finding the point at which the response and non-response histograms are most statistically separable, i.e. may comprise least similar signals. In other examples, the corrected response time may be found based on the statistical tests calculated for a range of response and non-response histograms for different corrected response times. Optionally the selected subset of frequencies comprises a plurality of frequency bands, wherein each band defines a range of frequencies, and the method further comprises measuring and recording one amplitude for each band of frequencies, the amplitude comprising an average or root mean square, RMS, across the range of frequencies of the frequency band. In this way, recording an amplitude for each frequency in a range is avoided, as the bands themselves contain a plurality of frequencies. Recording the amplitude of the band may comprise averaging, summing or otherwise calculating a representative amplitude of the plurality of frequencies in the band. In order to more accurately determine which part of the dynamic stimulus the subject is responding to, the method may further comprise monitoring a received stimulus at the subject, wherein the recording amplitudes of frequencies comprises recording amplitudes of frequencies in the received stimulus. Monitoring the received stimulus may be advantageous as the output device may cause changes in the amplitudes of different frequencies in the output stimulus, causing the subject to hear a different received stimulus to what was originally output. Further, if the output device is a significant distance from the subject, this may attenuate different frequencies differently depending on properties such as the size of the output device and the geometry of the space in which the method is being carried out. The monitoring of the received stimulus may comprise estimating the received stimulus. This may include estimating the effect of transducers used in the output device, such as speakers or headphones, on the dynamic stimulus. This estimation may comprise the use of a pre-programmed filter or series of filters, or a dedicated computer model of the effect of the output device on the dynamic stimulus. The effect may be consistent across a wide frequency range, or may be frequency dependent. Alternatively or in addition, the monitoring of the received stimulus may further comprise measuring a received stimulus at the subject. The received stimulus may be measured with a sound recording device, such as a microphone. The microphone may be positioned by the subject to measure the audio stimulus at the location of the subject. It is possible that the relative levels of different frequencies in a particular dynamic audio stimulus are not optimal for the assessment of the subject’s hearing. Therefore, the method may further comprise processing the dynamic audio stimulus to modify an intensity of a range of frequencies. The process of adjusting the intensity of different frequencies within an audio signal is commonly known as equalisation. The equalisation applied may be pre-specified or may be changed by the audiologist whilst the method is being carried out. The equalisation may increase or decrease the amplitude of different frequencies in the dynamic audio stimulus, in addition to the step of applying a filter to restrict the dynamic audio stimulus to a selected subset of the frequencies. Further, this equalisation may act in response to the monitoring of the received stimulus discussed above. The action of the equalisation may be to, in part or wholly, cancel out the differences between the output stimulus as intended to be output to the subject and the received stimulus. For example, if the output device is a speaker, and the speaker has a response in a particular frequency range that is erroneously intense, the equalisation may act to reduce the intensity of that frequency range such that the subject hears an undistorted, original stimulus. Alternatively or in addition, the method may further comprise applying compression to the dynamic audio stimulus. Compression, sometimes known as dynamic range compression, is an audio processing technique which reduces the intensity of high-intensity frequency components and increases the intensity of low-intensity frequency components. This reduces the dynamic range of a signal. Compression may be applied to a dynamic audio stimulus to bring the volume of different parts of the stimulus closer together, which may make the assessment of the subject’s hearing easier. The amount of compression applied, as well as the parameters of the compression, such as attack and release times and compression ratio, may be pre-specified or may be changed by the audiologist whilst the method is being carried out. A subject may be responsive to a dynamic audio stimulus filtered to relatively many frequency bands. It may then be important to determine more closely exactly what frequencies the subject is responding to. Therefore, the method may further comprise applying a filter to restrict the dynamic audio stimulus to a first selected subset of the frequencies and recording amplitudes of frequencies within the first subset of frequencies at a first measurement time based on a first response time; and changing the filter to restrict the dynamic audio stimulus to a second selected subset of the frequencies and recording amplitudes of frequencies within the second subset of frequencies at a second measurement time based on a second response time, wherein the second subset of frequencies are different to the first subset of frequencies. The second subset of frequencies may be a subset of the first subset of frequencies, or it may be a different subset of frequencies. In this way, it is possible to select different parts of the frequency spectrum to assess the subject’s hearing. When making multiple measurements and filtering the dynamic audio stimulus to distinct sets of frequencies, the reaction of the subject may occur at different amplitudes for the same frequencies. In this case, it may be advantageous to calculate minimum amplitudes that the subject can hear based on the recorded amplitudes. In this way, multiple measurements for the same frequency or frequency band can be combined into one estimation of the hearing threshold for said frequency or frequency band. In some cases, it may be advantageous to take the maximum amplitude, or to average the multiple amplitudes recorded. If there are multiple amplitudes recorded for a particular frequency or frequency band, it may be advantageous to estimate an error in the minimum amplitudes. This could comprise assuming the recorded amplitudes are normally distributed, or calculating a variance or expected value of the recorded amplitude for a particular frequency. This could comprise calculating an uncertainty in the recorded amplitudes or an error from an assumed “true” value. Calculating an error helps to discern where additional measurements may be required to give an accurate understanding of the subject’s hearing and therefore an accurate hearing threshold. When a wide range of frequencies are selected for filtering, as stated before, it can be useful to interpret which part of the dynamic audio stimulus the subject is responding to. Accordingly, the method may further comprise establishing which frequency or set of frequencies that the subject is responding to based on the recorded amplitudes. Establishing which set of frequencies the subject is responding to can be done by observing reactions to multiple overlapping frequency ranges and comparing the amplitude of each frequency band when the reaction is determined. In some examples, this may involve calculating a correlation between the amplitude of a particular frequency band and the likelihood of response. Optionally, the method may further comprise adjusting the filter to select which frequencies to restrict the dynamic audio stimulus to, based on the error in the minimum amplitudes. This reduces the input required from an operator, as well as reducing the domain knowledge required to optimally select the filtered frequencies. Additionally, this may reduce the number of measurements required as the selection can optimise which frequency ranges require additional measurements. A part of finding an accurate hearing threshold may be finding amplitudes to which the subject does not respond. Accordingly, the method may further comprise monitoring input data and determining a lack of response of the subject to the dynamic audio stimulus at a null response time, based on the input data; and recording amplitudes of frequencies within the subset of frequencies at a measurement time, the measurement time determined based on the response time or the null response time. This may comprise ascertaining an amplitude threshold for some or all of the frequencies, representing amplitudes that the subject cannot hear. The use of a dynamic audio stimulus, comprising multiple frequency components, can be advantageous in cases where a subject is not responsive to monotonic or non-dynamic audio stimuli. However, in some cases, the subject is only responsive, or is more responsive, to a particular stimulus, such as a particular piece of music or the theme tune of a favourite television show. Accordingly, the method may further comprise user selection of an audio file and outputting the dynamic audio stimulus based the selection. The selection may comprise provision of a computer-readable file, such as MP3, WAV, or FLAG audio. Alternatively, the selection may comprise provision of a means of accessing the audio file, such as a URL, a unique identifier, a link, or a search term. The selection may also take the form of a video file with embedded audio. As mentioned above, the response of the subject be delayed from when they heard the part of the dynamic audio stimulus to which they are responding. Accordingly, the measurement time may be determined by subtraction of a response lag time from the response time, wherein the response lag time may be adjustable based on the subject. The response lag time may compensate for the reaction time of the subject, as well as for a delay in determining a response and outputting the dynamic audio stimulus. For example, a young child or an elderly person may, on average, be perceived as responding to a dynamic audio stimulus more slowly than an adult between the ages of 18 and 50 years old. The response lag time allows for the recorded amplitudes to be the amplitudes at the time the subject actually responded, instead of the amplitudes a short time after the response. The response lag time may be determined by identifying the delay between the output of a stimulus the subject is known to respond to, and the subject exhibiting the response. This calibration process may be repeated to determine a reliable estimate of the response lag time. In challenging cases, the subject may not always exhibit a response to the dynamic audio stimulus despite hearing it. The method may implement further steps to introduce rewards of interest to keep the child engaged. In this way, the method may further comprise providing a visual stimulus to the subject when a response is determined. This may be an image or video displayed on a screen or projector in view of the subject, or a physical object revealed, lit up or displayed, such as a puppet or toy. The visual stimulus may be displayed automatically in response to the determined response. This may condition the subject to respond positively to the dynamic audio stimulus, as, when they respond, they are shown the visual stimulus. The method may further comprise providing an initial visual cue to the subject that the dynamic audio stimulus is playing, before a response has been registered. The visual cue may be displayed on a display. This may prompt the subject to focus. Selecting specific frequency ranges adds a level of complexity and can make it difficult to describe the hearing threshold. The frequencies selected for filtering may be grouped into frequency bands. The selected subset of frequencies may comprise a plurality of frequency bands, wherein each band defines a range of frequencies. The frequency bands may be of fixed or differing sizes. The bands may be a fixed width in frequency space, e.g. 100 Hz, or they may be relatively sized, such as each representing a successive octave. The size, e.g. one octave, and number of bands, e.g. seven bands, may be a fixed parameter, or it may be configurable. It may be advantageous to be able to view the frequency bands currently being filtered and not being filtered whilst carrying out the claimed method. Accordingly, the method may further comprise displaying a graph of frequency bands; and displaying on the graph the recorded amplitudes of frequencies. The recorded amplitudes may comprise a hearing threshold, and may additionally or alternatively comprise a null hearing threshold, denoting frequencies that the subject could not hear. The hearing threshold and null hearing threshold may be shown on the graph as a line or other indicator, which may be atop the graph of frequency bands. The graph of frequency bands may show their amplitude as the dynamic audio stimulus is output or may show their amplitudes at a particular time. Further, the graph may show which frequencies are currently selected for filtering. When selecting frequencies to filter, it can be difficult to ascertain which frequencies to filter next. To aid with this, the method may further comprise displaying a graph of selected subsets of frequencies previously selected for filtering the dynamic audio stimulus. This may comprise a separate display of previously selected subsets, or the information may be overlaid on a graph of frequency bands. The input data may comprise a manual input. The manual input may comprise an input from the subject, who may provide an input, such as a button press or a mouse click, when they have heard the dynamic audio stimulus. Additionally or alternatively, this manual input may be a button press or mouse click from an operator, who is observing the subject, the input provided when the operator believes the subject has responded. Alternatively, the input data may comprises measurement data relating to a measurement of the subject usable to determine a response. This may be advantageous in cases where the subject themselves is unable or unwilling to provide a manual input, and to save the labour of a third party providing the input themselves. This may decrease the time taken for the method. Further, this reduces the skilled labour of an operator who can notice subtle responses from the subject. Further still, responses determined from such input data may be more accurate and reliable than manual inputs, as they can take account of small changes in the measurements that may not be visible to the naked eye. This measurement data may be used as the sole source of a response, or it may be used in combination with other data, e.g. used when the subject does not manually indicate a response to the dynamic audio stimulus. This measurement data may comprise images or video of the subject. These images or video may be captured by a camera, such as a webcam, mobile phone, tablet or video recorder. This measurement data may be analysed to detect changes in posture, movements of the body or head, et cetera, which may correspond to a response to the dynamic audio stimulus. Alternatively or additionally, the measurement data may comprise data from sensors connected to the subject, such as a body-worn accelerometer or other movement sensor. Similarly, the measurement data may comprise measurements of the subject’s heart rate or brain activity. This measurement data may be analysed to detect changes in these quantities that may correspond to a response to the dynamic audio stimulus. Measurement data that doesn’t rely on physical movement of the body can be particularly advantageous in the case of a subject with a physical disability. Further, the measurement data may comprise audio recordings of the subject. These recordings may, in the case of a subject able and willing to participate in the measurement, comprise vocal affirmations of hearing or non-hearing at a particular time. In the case of a subject who is unwilling or unable to actively participate in the measurement, the audio recordings may be analysed to detect changes that may correspond to a response to the dynamic audio stimulus. A response may be determined from this measurement data when a change in the measurement data, e.g. corresponding to a change in body position or brain activity, corresponds to a change, e.g. an increased volume, in the dynamic audio stimulus. The amount of measurement data may be vast, which may be difficult to analyse analytically or with traditional computation. In order to traverse this difficulty, the method may further comprise inputting the measurement data into a trained machine learning model, the machine learning model trained to determine the presence of a user response based on input measurement data. The analysis of the measurement data may further comprise the calculation of an estimated confidence that the subject responded. This estimated confidence may be a confidence that the subject responded at all, or alternatively, an estimated confidence whether or not a known response was in fact in response to the stimulus or some other external factor. Further, as explained above, it can be difficult to analyse exactly which frequency components of a filtered dynamic audio stimulus a subject is responding to. In light of this challenge, the method may further comprise inputting the recorded amplitudes into the trained machine learning model, the machine learning model further trained to determine the hearing threshold of the subject based on the recorded amplitudes and input measurement data. Alternatively or in addition, the machine learning model may determine a predicted hearing threshold range, based on the recorded amplitudes and input measurement data. For example, the algorithm may find that the subject is more likely to respond to high-frequency sounds, or that the response lag time increases as the amplitude of the stimulus decreases. According to a second aspect, there is provided an audiometer, comprising an output for a dynamic audio stimulus, the dynamic audio stimulus comprising a plurality of different frequencies; and a processor, configured to: apply a filter to restrict the dynamic audio stimulus to a selected subset of the frequencies; monitor input data and determine a response of a subject at a response time based on the input data; and record amplitudes of frequencies within the subset of frequencies at a measurement time, the measurement time determined based on the response time. Optionally, the audiometer further comprises a sensor, configured to capture measurement data relating to the subject, and the processor is further configured to: input the measurement data into a trained machine learning model, the machine learning model trained to determine the presence of a response of the subject based on input measurement data; and / or input the recorded amplitudes into the trained machine learning model, the machine learning model further trained to determine the frequency bands the subject has responded to, based on the recorded amplitudes and input measurement data. According to a third aspect, there is provided a computer program product comprising instructions which, when executed by a processor, cause the processor to perform steps according to any other aspect. BRIEF DESCRIPTION OF DRAWINGS An example method and an example apparatus are described in detail herein with reference to the accompanying figures, in which: Figure 1 shows a schematic of example output from a computer-implemented method of audiometry, displaying an unfiltered audio stimulus; Figure 2 shows a schematic of example output from a computer-implemented method of audiometry, displaying a filtered audio stimulus; Figure 3 shows a schematic of example output from a computer-implemented method of audiometry, displaying a filtered audio stimulus with recorded amplitudes; Figure 4 shows a schematic of example output from a computer-implemented method of audiometry, displaying a filtered audio stimulus with recorded amplitudes; Figures 5 to 7 show a schematic of example output from a computer-implemented method of audiometry, displaying a filtered audio stimulus with recorded amplitudes for hearing and not hearing; and Figure 8 shows a schematic of the steps of an example method. DETAILED DESCRIPTION An example output display from a computer-implemented method of audiometry is generally illustrated at 1 in Figure 1. The display comprises a graph 20, which in turn comprises multiple bars 10, where each bar represents the current amplitude of a particular frequency band in the dynamic audio stimulus. The frequency bands are numbered with frequency labels 30, and their amplitude is labelled with amplitude labels 40. The graph 20 represents the amplitudes of different frequency components of the dynamic audio stimulus at a moment in time. In this example, the graph comprises bars 10, representing bands of frequencies. In other examples, the graph could comprise a histogram or line graph representing frequency bands of differing widths or single frequencies. In this example, the frequency bands represented by the bars 10 are each an octave wide, meaning that each successive frequency band is double the width of the previous. This is because an octave is defined as a doubling of frequency, typically measured in Hertz (Hz). Using bands an octave wide is idiomatic in the field of audiometry and fits well with the human perception of sounds. However, in other examples, bands of a fixed width in frequency space, e.g. 100 Hz wide, are used. In this example, the bars are labelled with frequency labels 30 that count sequential integers, but in other examples they may be labelled with their frequency ranges in Hertz. The length of a bar 10 ( / .e. its extension in the vertical direction) represents the amplitude of the frequencies in the frequency band, in this example measured in Decibels (db), as seen by the amplitude labels 40. In this example, higher-amplitude signals correspond to bars that extend further down the graph, an orientation which may be preferred by audiologists. However, in other examples, the bars could start at the bottom of the graph. In this example, the graph 20 updates in real time as the dynamic audio stimulus is played to the subject. In other examples, the graph may update with a delay or show the frequency components for a fixed part of the dynamic audio stimulus. An example method of audiometry corresponding to the claimed invention is demonstrated, by way of illustration, in Figures 2 to 7. The claimed invention allows amplitudes of frequencies or frequency bands that correspond to a response of a subject to be found. This corresponds to finding amplitudes of different frequency bands that the subject can hear. However, whilst ascertaining that a subject can hear a particular frequency at a particular amplitude is useful information, it is often useful to ascertain the lowest amplitude the subject can hear for a particular frequency. An example method of doing so is to determine a set of amplitudes that can be heard, decreasing the amplitude of the dynamic audio stimulus uniformly across all frequencies, then increasing it again until a response is registered. Repeating this process allows for optimizing the hearing threshold. The process can be further repeated, selecting different frequency bands for filtering. One example of this method is expounded below, with reference to Figures 2 to 7. In Figure 2, an example of the graph 20 is shown with a number of frequency bands selected for filtering, as shown by the filter window 50. In this example, the filter window 50 selects bands 3, 4 and 5 for filtering. In other examples, the filter window may select other bands, a discontinuous grouping of bands, a single band, or may comprise a frequency range (e.g. 25 to 100 Hz) that may or may not correspond to the widths of the bands. The selection of bands 10 that form the filter window 50 could be done by any means known in the art, including with a touch-screen display, a keyboard, a mouse, a dial or slider. In some examples, the selection of bands 10 for filtering may apply a band-pass filter, reducing the amplitude of non-selected bands to zero. In other examples, the amplitude of non-selected bands or non-selected frequencies may be reduced by a fixed amount, e.g. 10 db, or may be reduced proportional to the original amplitude, e.g. to 20% of the original amplitude. In this example, however, the amplitude of the non-selected bands is reduced in a non-uniform way. The closest non-selected bands 12 to the selected bands 10 are reduced in amplitude by a first percentage, say, 50% of the original amplitude. In other examples, they may be reduced by 20%, 40%, 60%, 80% or 100% of the original amplitude. The next closest bands 14 are reduced in amplitude by a second percentage, greater than the first percentage, say, 80% of the original amplitude. In other examples, they may be reduced by 20%, 40%, 60%, 80% or 100% of the original amplitude. This “tapering” of the non-selected bands around the selected bands allows the dynamic audio stimulus to retain some of its original character when filtered. It further decreases how harshly filtered the dynamic audio stimulus sounds. Both of these factors may improve the responsiveness of the subject to the filtered dynamic audio stimulus. The further a non-selected band is from the selected region the more intensely it is filtered and the less it contributes to the dynamic audio stimulus. In Figure 3, an example of the output display is shown after a response of the subject has been determined. In some examples, the response of the subject may be determined using machine learning, analysing the input data to determine if a response has taken place. In some examples, this involves the machine learning analysis of images or video of the subject, and / or measurements of heartrate or brain activity. The analysis may reveal a difference in this measurements from an expected value or baseline value, which can be correlated with a response to the dynamic audio stimulus. However, in this example, the input is provided by an operator who determines manually if there has been a response to the dynamic audio stimulus. The display is updated with a hearing threshold 60, indicating that the subject has reacted to the dynamic audio stimulus. The hearing threshold denotes the amplitude of the different frequency bands at the response time. As this is the first measurement, no comparison to previous measurements is performed. The response of the subject is determined at a response time, based on the input data. In this example, the measurement time, i.e. the time at which the amplitudes of the dynamic audio stimulus are measured, is distinct from the response time. It is calculated by the subtraction of a response lag time from the response time. In some examples, the response lag time can be adjusted based on how quickly the subject appears to be responding to known stimuli. The response lag time can be determined through playing stimuli at known sufficient amplitudes to get the subject to respond, then comparing the stimulus time to the response time. In other examples, the response lag time may be a fixed parameter. In other examples, the measurement time may be equal to the response time, or may be calculated as a function of the response time and / or other external factors, such as measurements of blood pressure or brain activity of the subject. In other examples, the measurement may not be carried out at a particular time. It may instead comprise taking the minimum, maximum or average amplitude of the frequency bands over a particular time window corresponding to the response time. In Figure 4, an example of the output display is shown after a first response of the subject has been determined and the amplitude of the dynamic audio stimulus has been subsequently reduced. In typical methods of audiometry known in the art, the audiologist establishes a response from the subject at a particular amplitude, then decreases the amplitude of the stimulus to below this threshold amplitude and increases it until a response is achieved again. The methods known in the art differ in many ways from the method presented here, for example, by their use of monotonic, non-dynamic audio stimuli and their manual detection of a response by the audiologist. As a first response of the subject has already been determined, the hearing threshold 60 is shown on the display. The frequency bands 10 show a decreased amplitude in comparison to the frequency bands of Figure 3. In Figure 5, an example of the output display is shown after a first response of the subject has been determined and the amplitude of the dynamic audio stimulus has been subsequently reduced. In this example, a lack of response of the subject to the dynamic audio stimulus has been determined. In some examples, this may be achieved by machine learning analysis of the input data, which could comprise images or video of the subject, or measurements of heartrate or brain activity. The lack of a significant change in these measurements from the baseline, or from the expected level, may be used to determine a lack of response. In this example, the lack of response is determined by an operator who then provides a suitable input. It should be understood that the lack of response can be determined in much the same fashion as the response, and that description of determination of a response could be applied, wherever possible, to determination of a lack of response. The display is updated with a null hearing threshold 70, indicating that the subject has not reacted to the dynamic audio stimulus. This implies that the subject did not hear the dynamic audio stimulus. The null hearing threshold denotes the amplitude of the different frequency bands at the null response time. As this is the first null response measurement, no comparison to previous null response measurements is performed. The comparison to previous null response measurements is carried out in an analogous way to the comparison to previous response measurements described below for Figure 7. As for the response measurement presented above, there is a distinction between the measurement time, i.e. the time at which the amplitudes of the dynamic audio stimulus are measured, and the null response time. The discussion above relating to the response time should be understood to apply, where appropriate, to the null response time. In this example, the measurement time is calculated by subtraction of a response lag time from the null response time. In other examples, a different lag time may be used, or no lag time at all may be used. This is because the individual amplitudes of the bands at any particular time is less important for a null response than a response. In other examples, the measurement may not be carried out at a particular time. It may instead comprise taking the minimum, maximum or average amplitude of the frequency bands over a particular time window corresponding to the null response time. In Figure 6, an example of the output display is shown after a first response of the subject has been determined, and a subsequent null response of the subject has been determined. In this example, the amplitude of the dynamic audio stimulus has been subsequently increased. In doing so, it becomes possible to expose the subject to amplitudes for particular frequency bands which are lesser than the current hearing threshold, but greater than the current null hearing threshold. For the sake of simplicity, Figure 6 shows an example of the output display the instant before a response to the dynamic audio stimulus is determined. One of the selected bands 18 has an amplitude below the current hearing threshold for that band. If the subject responds to the dynamic audio stimulus at this time, it implies that they may be able to hear this frequency band at a lower amplitude than previous measurements implied. This requires updating of the hearing threshold to denote that, for this frequency band, there is a new lower threshold for hearing. This updating is seen in Figure 7, where a response to the dynamic audio stimulus of Figure 6 has been determined. The updated hearing threshold 62 is shown. In comparison to Figure 6, the updated hearing threshold 62 has a lower amplitude recorded for one of the bands. This is illustrative of the iterative method of audiometry presented, where the hearing threshold and the null hearing threshold are bought closer together via a series of sequential responses and null responses at different amplitudes of the dynamic audio stimulus. In the examples presented above, the input data is input provided by an operator who determines manually if there has been a response to the dynamic audio stimulus. However, in some examples, the input data comprises measurement data relating to a measurement of the subject usable to determine a response. In some of these examples, the measurement data comprises images and / or video of the subject whilst the dynamic audio stimulus is played. This measurement is analysed using machine learning. In these examples, the machine learning algorithms are of an image classifier type, capable of identifying elements in an image with a given certainty (a function of how closely the element matches the elements in the training data, as a percentage, for example). These algorithms may comprise neural networks or convolutional neural networks. These algorithms typically require a large corpus of annotated training data, where pairs of images and the desired output from the algorithm for that image are given, and statistical mappings between the training data images and the desired outputs are made. Once this training process is complete, the machine learning algorithms are trained and no longer need access to the corpus of training data. The training corpus in this example is a set of annotated images, with annotations corresponding to whether or not the subject is responding to the dynamic audio stimulus. In other cases, the annotations may correspond to what actions the subject is currently performing, or the subject’s perceived emotional state. In some examples, the subject may exhibit the traits of a response, e.g. physical movement, but the response is not due to the dynamic audio stimulus. Instead, the response is due to some external factor, such as a noise from outside the room. Detection of such false responses may be performed by a machine learning algorithm trained to do so, or the false responses may be noted by the operator whilst the method is being performed. Such a machine learning algorithm can be trained on a corpus of annotated data corresponding to “real” and “false” responses to dynamic audio stimuli. In other examples, machine learning analysis is used to determine which frequency component of the dynamic audio stimulus the subject has responded to. In such examples, the machine learning analysis may be carried out using neural networks, convolutional neural networks or transformers. Alternatively, in other examples, the correlation between amplitudes of frequency components and the likelihood of a response is calculated using statistical techniques. This includes the calculation of the likelihood of a particular frequency component causing a response, using a plurality of recorded amplitudes. In some examples, each correct response to an dynamic audio stimulus results in a score being allocated to the subject. This scoring system provides a quantitative measure of the subject’s performance, which can be used to track progress over time, compare performance across different subjects, or evaluate the effectiveness of different stimuli or testing protocols. In some examples, the score is calculated based on one or more of: the speed of the subject’s response, the accuracy of the subject’s response, the difficulty of the stimulus, the subject’s previous performance. This scoring system can be implemented using standard statistical methods. In some examples, the scoring system is used to attach a confidence metric to the hearing threshold. In some examples, the score allocated to the user is based on the speed that the subject responds to the dynamic audio stimulus. In Figure 8, an example schematic is shown, detailing the steps of one example process. The process begins with an output step S100, where the dynamic audio stimulus, comprising a plurality of different frequencies, is output to the subject. A filter is applied to the dynamic audio stimulus, as it is output, in a filtering step S102. Input data is monitored and a response of the subject is determined based on that input data in an input monitoring step S104. Then, amplitudes of frequencies within the subset of frequencies filtered to are measured and recorded at a measurement time in the recording step S106. The process can be repeated in order to assess the hearing range of the subject in different frequency ranges or with greater precision. The discussion and exposition given above relating to other examples should be understood as applying, wherever possible, to the steps of this example process also.

Claims

1. A computer-implemented method of audiometry, the method comprising: outputting a dynamic audio stimulus to a subject, the dynamic audio stimulus comprising a plurality of different frequencies;applying a filter to the dynamic audio stimulus as it is output, to restrict the dynamic audio stimulus to a selected subset of the frequencies;monitoring input data and determining a response of the subject at a response time based on the input data;recording an amplitude of one or more frequencies within the subset of frequencies at a measurement time, the measurement time determined based on the response time.

2. The method of claim 1, further comprising determining a hearing threshold based on the recorded amplitude.

3. The method of claim 2 further comprising: determining a plurality of responses at different measurement times during output of the filtered dynamic audio stimulus,for each determined response, recording an amplitude of one or more frequencies within the subset of frequencies at a measurement time, the measurement time at or before the response time;determining a hearing threshold for the subset of frequencies based on the amplitudes recorded across the plurality of responses.

4. The method of claim 3 further comprising: monitoring input data while applying a plurality of different filters to the dynamic audio stimulus, each filter restricting the dynamic audio stimulus to a different subset of frequencies;determining a plurality of responses at different measurement times while applying each filter;recording an amplitude of one or more frequencies within the subset of frequencies for each of the plurality of responses determined for each of the plurality of filters;determining a hearing threshold for each subset of frequencies based on the recorded amplitudes.

5. The method of any preceding claim, wherein the amplitudes of the different frequencies of the dynamic audio stimulus vary with time, and the method comprises:measuring and recording an amplitude of one or more frequencies within the subset of frequencies at a plurality of measurement times during a response measurement window, the response measurement window determined based on the response time.

6. The method of claim 5, further comprising:measuring and recording an amplitude of one or more frequencies within the subset of frequencies at a plurality of non-response measurement times during a non-response measurement window, the non-response measurement window determined based on the response time; anddetermining a hearing threshold based on comparing the amplitudes recorded during the response measurement window and the non-response measurement window.

7. The method of claim 4, further comprising:measuring and recording the amplitude of one or more frequencies within the subset of frequencies at a plurality of non-response measurement times during a non-response measurement window, the non-response measurement window determined based on the response time; anddetermining a hearing threshold range based on the amplitude recorded during the response measurement window and the amplitude recorded during the non-response measurement window.

8. The method of claim 6 or claim 7, further comprising recording the maximum or average amplitude across the plurality of measurement times.

9. The method of any preceding claim, wherein the selected subset of frequencies comprises a plurality of frequency bands, wherein each band defines a range of frequencies, the method comprising:measuring and recording one amplitude for each band of frequencies, the amplitude comprising an average or root mean square, RMS, across the range of frequencies of the frequency band.

10. The method of any preceding claim, further comprising monitoring the dynamic audio stimulus as it is received at the subject, and wherein the recording amplitudes of frequencies comprises recording amplitudes of frequencies in the received dynamic audio stimulus.

11. The method of claim 10, wherein monitoring the received stimulus at the subject comprises estimating the received dynamic audio stimulus at the subject based on the output stimulus or measuring a received stimulus at the subject with a microphone.

12. The method of any preceding claim, further comprising:applying a filter to restrict the dynamic audio stimulus to a first selected subset of the frequencies and recording amplitudes of frequencies within the first subset of frequencies at a first measurement time based on a first response time; andchanging the filter to restrict the dynamic audio stimulus to a second selected subset of the frequencies and recording amplitudes of frequencies within the second subset of frequencies at a second measurement time based on a second response time, wherein the second subset of frequencies are different to the first subset of frequencies.

13. The method of claim 12, further comprising estimating an error in the minimum amplitudes; and adjusting the filter to select which frequencies to restrict the dynamic audio stimulus to, based on the error in the minimum amplitudes.

14. The method of any of claims 15 to 17, further comprising establishing which frequency or set of frequencies that the subject is responding to based on the recorded amplitudes.

15. The method of any preceding claim, further comprising:monitoring input data and determining a lack of response of the subject to the dynamic audio stimulus at a null response time, based on the input data; andrecording amplitudes of frequencies within the subset of frequencies at a measurement time, the measurement time determined based on the response time or the null response time.

16. The method of any preceding claim, further comprising receiving a user selection of an audio file and outputting the dynamic audio stimulus based the selection.

17. The method of any preceding claim, wherein the measurement time is determined by subtraction of a response lag time from the response time, wherein the response lag time is adjustable based on the subject.

18. The method of any preceding claim, further comprising providing a visual stimulus to the subject when a response is determined.

19. The method of any preceding claim, wherein the input data comprises measurement data relating to a measurement of the subject usable to determine a response, wherein the measurement data comprises images or video of the subject.

20. The method of any preceding claim, wherein the input data comprises measurement data relating to a measurement of the subject usable to determine a response, wherein the measurement data comprises heart rate or brain activity measurements.

21. The method of claim 19 or 20, wherein the method comprises inputting the measurement data into a trained machine learning model, the machine learning model trained to determine the presence of a response of the subject based on input measurement data.

22. The method of any preceding claim, wherein the method further comprises inputting the recorded amplitudes into a trained machine learningmodel, the machine learning model trained to determine a hearing threshold, based on the recorded amplitudes and input measurement data.

23. An audiometer, comprising:an audio output; anda processor, configured to:output a dynamic audio stimulus to a subject with the audio output, the dynamic audio stimulus comprising a plurality of different frequenciesapply a filter to restrict the dynamic audio stimulus to a selected subset of the frequencies;monitor input data and determine a response of a subject at a response time based on the input data; andrecord amplitudes of frequencies within the subset of frequencies at a measurement time, the measurement time determined based on the response time.

24. The audiometer of claim 23, further comprising a sensor configured to capture measurement data relating to the subject, wherein the processor is further configured to:input the measurement data into a trained machine learning model, the machine learning model trained to determine the presence of a response of the subject based on input measurement data; and / orinput the recorded amplitudes into the trained machine learning model, the machine learning model further trained to determine the frequency bands the subject has responded to, based on the recorded amplitudes and input measurement data.

25. A computer program product comprising instructions which, when executed by a processor, cause the processor to perform steps according to any one of claims 1 to 22.AMENDMENTS TO THE CLAIMS HAVE BEEN FILED AS FOLLOWS:-18 09 24CLAIMS1. A computer-implemented method of audiometry, the method comprising: outputting a dynamic audio stimulus to a subject, the dynamic audio stimulus comprising a plurality of different frequencies, wherein the amplitudes of 5 the different frequencies of the dynamic audio stimulus vary with time,applying a filter to the dynamic audio stimulus as it is output, to restrict the dynamic audio stimulus to a selected subset of the frequencies:monitoring input data and determining a response of the subject at a response time based on the input data;10 measuring and recording an amplitude of one or more frequencies withinthe subset of frequencies at a plurality of measurement times during a response measurement window, the response measurement window determined based on the response time;measuring and recording the amplitude of one or more frequencies within 15 the subset of frequencies at a plurality of non-response measurement times during a non-response measurement window, the non-response measurement window determined based on the response time, in which the subject does not respond to the dynamic audio stimulus; anddetermining a hearing threshold range based on the amplitudes recorded 20 during the response measurement window and the amplitudes recorded during the non-response measurement window.

2. The method of claim 1, wherein the hearing threshold range comprises a an upper bound and a lower bound, wherein the upper bound is determined from the amplitudes recorded during the response measurement window, and the lower 25 bound is determined from the amplitudes recorded during the non-response measurement window, wherein the upper bound and lower bound correspond to an estimated minimum and maximum amplitude for the hearing threshold.

3. The method of claim 1 or claim 2, further comprising determining a hearing threshold based on the recorded amplitude.determining a plurality of responses at different measurement times during output of the filtered dynamic audio stimulus,for each determined response, recording an amplitude of one or more frequencies within the subset of frequencies at a measurement time, the 5 measurement time at or before the response time;determining a hearing threshold for the subset of frequencies based on the amplitudes recorded across the plurality of responses.

5. The method of claim 4 further comprising:monitoring input data while applying a plurality of different filters to the10 dynamic audio stimulus, each filter restricting the dynamic audio stimulus to a different subset of frequencies;determining a plurality of responses at different measurement times while applying each filter;CMrecording an amplitude of one or more frequencies within the subset of15 frequencies for each of the plurality of responses determined for each of the plurality of filters;determining a hearing threshold for each subset of frequencies based on the recorded amplitudes.

6. The method of claim 1, further comprising:20 determining a hearing threshold based on comparing the amplitudesrecorded during the response measurement window and the non-response measurement window.

7. The method of any preceding claim, further comprising recording the maximum or average amplitude across the plurality of measurement times.25 8. The method of any preceding claim, wherein the selected subset offrequencies comprises a plurality of frequency bands, wherein each band defines a range of frequencies, the method comprising:measuring and recording one amplitude for each band of frequencies, the amplitude comprising an average or root mean square, RMS, across the range of 30 frequencies of the frequency band.18 09 249. The method of any preceding claim, further comprising monitoring the dynamic audio stimulus as it is received at the subject, and wherein the recording amplitudes of frequencies comprises recording amplitudes of frequencies in the received dynamic audio stimulus.5 10. The method of claim 9, wherein monitoring the received stimulus at thesubject comprises estimating the received dynamic audio stimulus at the subject based on the output stimulus or measuring a received stimulus at the subject with a microphone.

11. The method of any preceding claim, further comprising:10 applying a filter to restrict the dynamic audio stimulus to a first selectedsubset of the frequencies and recording amplitudes of frequencies within the first subset of frequencies at a first measurement time based on a first response time; andchanging the filter to restrict the dynamic audio stimulus to a second 15 selected subset of the frequencies and recording amplitudes of frequencies within the second subset of frequencies at a second measurement time based on a second response time, wherein the second subset of frequencies are different to the first subset of frequencies.

12. The method of claim 11, further comprising estimating an error in the 20 minimum amplitudes; and adjusting the filter to select which frequencies to restrict the dynamic audio stimulus to, based on the error in the minimum amplitudes.

13. The method of any preceding claim, further comprising establishing which frequency or set of frequencies that the subject is responding to based on the recorded amplitudes.25 14. The method of any preceding claim, further comprising:monitoring input data and determining a lack of response of the subject to the dynamic audio stimulus at a null response time, based on the input data; and18 09 24recording amplitudes of frequencies within the subset of frequencies at a measurement time, the measurement time determined based on the response time or the null response time.

15. The method of any preceding claim, further comprising receiving a user 5 selection of an audio file and outputting the dynamic audio stimulus based the selection.

16. The method of any preceding claim, wherein the measurement time is determined by subtraction of a response lag time from the response time, wherein the response lag time is adjustable based on the subject.10 17. The method of any preceding claim, further comprising providing a visualstimulus to the subject when a response is determined.

18. The method of any preceding claim, wherein the input data comprises measurement data relating to a measurement of the subject usable to determine a response, wherein the measurement data comprises images or video of the 15 subject.

13. The method of any preceding claim, wherein the input data comprises measurement data relating to a measurement of the subject usable to determine a response, wherein the measurement data comprises heart rate or brain activity measurements.20 20. The method of claim 18 or 19, wherein the method comprises inputtingthe measurement data into a trained machine learning model, the machine learning model trained to determine the presence of a response of the subject based on input measurement data.

21. The method of any preceding claim, wherein the method further 25 comprises inputting the recorded amplitudes into a trained machine learning model, the machine learning model trained to determine a hearing threshold, based on the recorded amplitudes and input measurement data.an audio output, anda processor, configured to:output a dynamic audio stimulus to a subject with the audio output, the dynamic audio stimulus comprising a plurality of different5 frequencies, wherein the amplitudes of the different frequencies of thedynamic audio stimulus vary with time;apply a filter to restrict the dynamic audio stimulus to a selected subset of the frequencies:monitor input data and determine a response of a subject at a10 response time based on the input data;measure and record amplitudes of one or more frequencies within the subset of frequencies at a plurality of measurement times during a response measurement window, the response measurement window determined based on the response time;15 measure and record the amplitude of one or more frequencieswithin the subset of frequencies at a plurality of non-response measurement times during a non-response measurement window, the non-response measurement window determined based on the response time, in which the subject does not respond to the dynamic audio20 stimulus; anddetermine a hearing threshold range based on the amplitude recorded during the response measurement window and the amplitude recorded during the non-response measurement window.

23. The audiometer of claim 22, further comprising a sensor configured to 25 capture measurement data relating to the subject, wherein the processor is further configured to;input the measurement data into a trained machine learning model, the machine learning model trained to determine the presence of a response of the subject based on input measurement data; and / or30 input the recorded amplitudes into the trained machine learning model,the machine learning model further trained to determine the frequency bands theCMsubject has responded to, based on the recorded amplitudes and input measurement data.

24. A computer program product comprising instructions which, when executed by a processor, cause the processor to perform steps according to any 5 one of ciaims 1 to 21.

Images