WO2016154139A1 - Sound-based spirometric devices, systems, and methods using audio data transmitted over a voice communication channel - Google Patents

Abstract

Examples described herein include a call-in service that measures lung function using any phone without the need for an application running on the phone itself. Examples described herein may enable the user to estimate their lung function by performing a spirometry effort while a communication device collects audio data relating to the effort. The collected audio data may be transmitted to a server over a standard voice channel. The server may calculate clinically relevant lung function measures, which can be reported back to the participants using audio during the call or over SMS.

Description

SOUND-BASED SPIROMETRIC DEVICES, SYSTEMS, AND METHODS USING AUDIO DATA TRANSMITTED OVER A VOICE COMMUNICATION CHANNEL

CROSS-REFERENCE TO RELATED APPLICATIONS)

[001] This application claims the benefit under 35 U.S.C. 119 of the earlier filing date of

U.S. Provisional Application 62/136,095, filed March 20, 2015, the entirety of which is hereby incorporated by reference in its entirety for any purpose.

TECHNICAL FIELD

[002] Examples described herein relate to testing lung function; examples of evaluating lung function through a phone call are described.

BACKGROUND

[003] Spirometry is a mainstay for measuring lung function and is central for the diagnosis of chronic lung impairments, such as asthma, chronic obstructive pulmonary disease (COPD), and cystic fibrosis. During spirometry tests, participants generally forcefully exhale all the air out of their lungs. The spirometer measures the instantaneous flow and cumulative volume of exhaled air. It then calculates various lung function measures, such as levels of obstruction or restriction, to help diagnose and manage various pulmonary conditions.

[004] Spirometry is a widely employed pulmonary function test. There are many different types of spirometers available, ranging from big, clinical spirometers to portable, home spirometers. Their cost also generally varies from $1,000 USD to $5,000 USD. During a spirometry test, the subject generally takes the deepest breath possible and then exhales with maximum force for as long as possible. The spirometer measures the amount and speed of airflow and calculates various indications of lung function on the basis of the test. Four example indications of lung function are:

[005] (l)Forced Vital Capacity (FVC): The total volume of air expelled during the expiration,

[006] (2) Forced Expiratory Volume in one second (FEV1): The volume of air expelled in the first second of expiration,

[007] (3) FEV 1/FVC (FEV 1 %) : Ratio of FEV 1 and FVC, and [008] (4) Peak Expiratory Flow (PEF): Maximum expiratory flow rate reached during the test.

Θ09] Generally a healthy individual's lung function measures are at least 80% of the values predicted based on their age, height, and gender. Abnormal values of FEV1% may be (expressed as a percent of predicted value):

[010] · Mild Lung Dysfunction: 60-79%,

[011] · Moderate Lung Dysfunction: 40-59%, and

[012] · Severe Lung Dysfunction: below 40%.

[013] Apart from the numerical measures, spirometers may also generate Flow vs. Time

(FT), Flow vs. Volume (FV), and Volume vs. Time (VT) plots.

[014] Figure 1 is an example flow vs. volume plot for normal, obstructive, and restrictive cases. The plot 102 includes normal case 104, obstructive case 106, and restrictive case 108. The plot 102 shows flow on the y-axis in liters/second, and volume on the x-axis in liters. The plot 102 also illustrates the PEF indication of lung function (e.g. shown as a peak amplitude of the plot in any given case), the FEVl indication (e.g. shown as a volume reached on certain cases after one second), and the FVC indication (e.g. shown as total volume expelled in certain cases). In a healthy individual, the normal case 104 illustrates that a descending limb of the FV plot is almost a straight line. As obstruction to the airflow increases, the flow rate decreases faster than exponentially after reaching its maximum value (PEF). Therefore, it attains a curved or "scooped" slope in the obstractive case 106. For an individual suffering from a restrictive lung disease, such as cystic fibrosis, the respiratory muscles weaken and the patient's lung capacity (FVC) decreases as shown in the restrictive case 108.

SUMMARY

[015] Examples of methods are described herein. Some example methods include receiving, from a communication device, through a voice communication channel, audio data obtained by a microphone of the communication device during inhalation and exhalation of a user, and transmitting, to the communication device, an indication of lung function of the user based on the audio data. [016] In some examples, the audio data is compressed by the voice communication channel.

[017] In some examples, the audio data is compressed by the voice communication channel using linear predictive coding,

[018] In some examples, the voice communication channel is a GSM channel.

[019] In some examples, the indication of lung function is provided using audio or short messaging sen' ice.

[020] Some example methods further include extracting features from the audio data and determining the indication of lung function based, at least in part, on the features.

[021] In some examples, wherein the features include linear predictive coding features, signal envelope, and resonance tracking features.

[022] In some examples, determining the indication of lung function includes performing regression on the features.

[023] In some examples, performing regression includes performing a linear regression, a least-angle regression, an elastic net regression, and a kN regression .

[024] In some examples, determining the indication of lung function includes taking a median of regressions performed.

[025] Some examples further include estimating channel state information of the voice communication channel, and determining the indication of lung function based, at least in part, on the channel state information.

[026] In some examples, the indication of lung function includes forced vital capacity, forced expiratory volume in one second, peak expiratory flow, ratios of those, or combinations thereof.

[027] Some example methods include holding a communication device at arm's length, fully inhaling and exhaling, transmitting audio data collected by the communication device during said fully inhaling and exhaling to a remote server through a voice communication channel, and receiving an indication of lung function corresponding to the audio data at the communication device through audio or short messaging service communication.

[028] In some examples, the voice communication channel is a GSM channel.

[029] Some example methods further include dialing, with the communication device, a predetermined phone number for a call-in service prior to said transmitting audio data. [030] In some examples, fully inhaling and exhaling includes performing a spirometry effort.

[031] In some examples, the indication of lung function is based, at least in part, on features extracted from the audio data.

[032J In some examples, the indication of iung function is based, at least in part, on a linear regression, a least-angle regression, an elastic net regression, and a kN regression of the features.

[033] Some examples further include transmitting a set of pilot tones through the voice communication channel.

[034] Examples of systems are described herein. Some example systems include a communication device, and a computing system, the computing system including at least one processing unit and computer readable media encoded with instructions which, when executed by the at least one processing unit, cause the computing system to perform, actions including: receiving, from the communication device, through a voice communication channel, audio data obtained by a microphone of the communication device during inhalation and exhalation of a user, and transmitting, to the communication device, an indication of lung function of the user based on the audio data.

Θ35] In some examples, the computing system may be a first computing system and the communication device may be configured to provide the audio data to a second computing system through the voice communication channel. In some examples the second computing system is configured to provide the audio data to the first computing system,

[036] In some examples, the second computing system includes a voice server of a telecommunications network.

[037] In some examples, the indication of lung function is provided using audio or short messaging service.

[038] In some examples, the actions further include extracting features from the audio data and determining the indication of lung function based, at least in part, on the features.

[039] In some examples, the actions further include performing regression on the features including performing a linear regression, a least-angle regression, an elastic net regression, and a kNN regression. [040] In some examples, the actions further include estimating channel state information of the voice communication channel, and determining the indication of lung function based, at least in part, on the channel state information,

BRIEF DESCRIPTION OF THE DRAWINGS

[041] FIG. 1 is an example flow vs. volume plot for normal, obstructive, and restrictive cases.

[Θ42] FIG. 2 is a schematic illustration of a system arranged in accordance with examples described herein.

[043] FIG. 3 is a flowchart of a method arranged in accordance with examples described herein.

[044] FIG. 4 is a flowchart of a method arranged in accordance with examples described herein.

[045] FIG. 5 is a flowchart of an example method arranged in accordance with examples described herein.

DETAILED DESCRIPTION

[046] Certain details are set forth below to provide a sufficient understanding of embodiments of the invention. However, it will be clear to one skilled in the art that embodiments of the invention may be practiced without various of these particular details. In some instances, well-known circuits, control signals, telecommunication components, communication device components, timing protocols, and software operations have not been shown in detail in order to avoid unnecessarily obscuring the described embodiments of the invention.

[047] Local smartphone applications may be provided that sample the smartphone microphone and send digital sound generated by the patient's vocal tracts as well as relevant features captured during the maneuver to a central server. The server may then calculate the expiratory flo rate using a physiological model of the vocal tract and a model of the reverberation of sound around the user's head. Some example applications are described, for example, in co-pending U.S. Application Serial Number 14/400,064, filed May 10, 2013 entitled ''Sound-based spirometric devices, systems, and methods ' which application is hereby incorporated by reference in its entirety for any purpose, and U .S. Provisional Application No. 61/645,176, filed May 10, 2012, which application is incorporated herein by reference in its entirety for any purpose.

I Use of smartphone applications for this purpose, however, may require a custom application to be installed on the smartphone to sample high quality audio data and to extract features for further processing. Unfortunately, modern smartphone devices are still not common in many regions of the world. Even as the penetration of smartphones increases, the diversity of phone manufactures and models make it challenging to manage custom applications for every type of mobile phone that may be in use. Feature phones do not have a consistent programming interface and the variability in microphones and sampling rates can be significant. Considering that a major portion of the population suffering from lung impairments lives in low resource environments, it would be desirable for phone-based spirometers to work on a greater number of, or all, mobile phones, not just programmable smartphones.

Examples described herein include a call-in service that measures lung function using any phone without the need for an application running on the phone itself. Examples described herein may enable the user to estimate their lung function by holding the phone at arm's length, fully inhaling, and forcefully exhaling until ail the air is expelled. The collected audio data may be transmitted to a server over a standard voice channel. The server may calculate clinically relevant lung function measures, which can be reported back to the participants using audio during the call or over SMS. The ability to use a server to analyze audio data transmitted from any mobile phone, be it a wired phone, feature phone or smartphone, may eliminate or reduce a need to develop a specialized application for every phone platform. This approach may make the client device, channel, and app agnostic and keep the intelligence on the central server.

I Generally, examples described herein utilize a microphone on a user's communication device to obtain audio data relating to a user's spirometry maneuver (e.g. full and/or forced inhalation and exhalation). No intermediate device or transducer may be used in some examples - e.g. the communication device may capture audio data resulting from an unaided inhalation and exhalation. In some examples, however, a whistle or other device may be used to convert the sounds of the spirometry maneuver into tones that are then captured by the microphone of the communication device. Such a whistle or other device may be advantageous in users having a very low flow rate, such that the sound generated by the vocal tracts might be negligible. Such patients may usually have very low lung function (e.g. FEV1% < 50%) and are generally willing to carry an additional lung function measurement device with them. The whistle or other device may be provided as an attachment or accessory for a user's communication device.

] Figure 2 is a schematic illustration of a sy stem arranged in accordance with examples described herein. The system 214 includes a communication device 210, voice communication channel 218, optional voice server 212, optional connection 220, and computing system 202. The computing system 202 includes one or more processing unit(s) 204 and memory 206 which may encode executable instructions for measuring lung function 208.

] Generally, the communication device 210 may record audio data from a user's spirometry effort. The audio data may be provided over the voice communication channel 218 to the computing system. 202. In some examples, the audio data may be provided to the computing system 202 through use of one or more voice servers, such as the voice server 212. The voice communication channel 218 may compress and/or encode the audio data, and accordingly the methods and algorithms employed by the computing system 202 (e.g. the instructions provided by the executable instructions for measuring lung function 208) are sufficiently robust in some examples to provide indications of lung function even in the presence of compression or other loss through the voice communication channel 218.

] The communication device 210 may be implemented using generally any device suitable for placing a call over the voice communication channel 218. Examples include, but are not limited to, cell phones, smart phones, feature phones, mobile phones, landline phones, satellite phones, Internet phones, and combinations thereof. Other devices capable of communication over the voice communication channel 218 may be used including tablets, computers, servers, laptops, automobiles, appliances, wearable devices, virtual and augmented reality devices, and combinations thereof. While a single communication device 210 is shown in Figure 2, any number of communication devices may provide audio data to the computing system 202 over the voice communication channel 218 and/or other voice communication channels.

] The communication device 210 generally includes a microphone to transduce analog sound waves into electronic signals. The sounds produced by a user performing a spirometry effort (e.g. full inhalation and exhalation and/or inhalation followed by forced exhalation) may be received by the microphone and converted into electronic signals (e.g. currents, voltages). The communication device 210 generally further includes a transmitter for transmitting audio data (e.g. data indicative of the electronic signals provided by the microphone) over the voice communication channel 218.

] Accordingly, the communication device 210 may obtain audio data during a spirometry effort of a user by capturing audio data during the spirometry effort. The spirometry effort may in some examples be unaided by any further audio-generating apparatus (e.g. a mobile phone itself may be used to pick up the sounds of a user performing a spirometry effort). However, in other examples the user may utilize a device, e.g. a whistle or other tone-producing device that may convert flow during a spirometry effort into audio tones that are then recorded by the communication device.

] The voice communication channel 218 may be implemented using generally any communication channel used to carry voice communications. Communication channels used in telephony networks may be used to implement the voice communication channel 218. Examples include, but are not limited to, GSM, UMTS, LTE, or PSTN communication channels. While voice communication channels associated with telephony networks may be used to implement the voice communication channel 218, in some examples, Wi-Fi may additionally or instead be used to transmit data between the communication device 210 and the computing system 202,

] The computing system 202 may be implemented using, for example, a server (e.g. a remote server), a controller, a microcontroller, a desktop, a laptop, a tablet, or a smartphone. The computing system 202 may include one or more processing unit(s) 204 (e.g. processors), circuitry, firmware) and computer readable media (e.g. Memory 206) which may be encoded with executable instructions for measuring lung function 208. For example, the computing system 202 may be programmed with the executable instructions for measuring lung function 208.

[058] The computer readable media may be implemented using generally any electronic memory including, but not limited to, Flash, SSD, RAM, ROM, or disk drives. It should be understood that the arrangement of the computing system 202 is quite flexible. For example, the processing unit(s) 204 and memory 206 may in some examples be distributed and in electronic communication. The memory 206 (or other computer readable media included in the computing system 202) may store data used during the lung function analysis described herein, including, but not limited to, lung function parameters.

[059] The computing system. 202, may include various other computing system, components, including but not limited to input and/or output device(s) such as display(s), keyboard(s), mice, touchscreen(s), virtual or augmented reality displays, or the like.

[060] The computing system 202 may include a receiver to receive audio data from the communication device 210 over the voice communication channel 218. In some example, the computing system 202 may receive the audio data from the voice communication channel 218 directly.

[061] The computing system 202, may include a transmitter to transmit indications of lung function back to the communication device 210 through the voice communication channel 218. The indications of lung function may be provided, for example in audio or using short messaging sen' ice (SMS).

[062] In some examples, an optional voice server 212 may mediate communication from the voice communication channel 218 to the computing system 202 over connection 220. The voice server 212 may be a voice server of a telecommunication network in some examples. The connection 220 may be implemented, for example, using a Wi-Fi, Internet, or wired connection. In some examples, another voice communication channel (e.g. using a telephony network) may be used to implement the connection 220.

[063] During operation, the computing system 202 may receive audio data obtained by a microphone of the communication device 2, 10 during performance of a spirometry effort by a user (e.g. an inhalation and exhalation). The computing system 202 may process the audio data using the executable instructions for measuring lung function 208 to determine one or more indicators of lung function (e.g. PEF, FEV 1, FVC) based on the audio data. The computing system 202 may provide one or more indications of lung function back to the communication device 210 over the voice communication channel 218 (in some examples indirectly through the voice server 212 or other intermediate server).

[064J The computing system 202 may generally be configured (e .g. programmed) to implement machine learning techniques to determine indications of lung function based on received audio data,. The executable instructions for measuring lung function 208 may- include instructions for performing machine learning techniques described herein.

[065] The executable instructions for measuring lung function 208 may, when executed by the processing unit(s) 204, cause the computing system 202 to perform functions described herein, including the receipt of audio data from the communication device 210 and transmission of one or more indicators of lung function to the communication device 210.

[066] In some examples, the executable instructions for measuring lung function 208 including instructions for extracting features from the audio data and determining the indication of lung function based, at least in part, on the features. In some examples, the executable instructions for measuring lung function 208 include instructions for performing regression on the features including performing a linear regression, a least-angle regression, an elastic net regression, and a kNN regression .

[067] In some examples, the instructions further include instructions for estimating channel state information of the voice communication channel 2, 18 and determining the indication of lung function based, at least in part, on the channel state information,

[068] The method 300 includes, at block 302, receiving, from a communication device, through a communication channel, audio data obtained by a microphone of the communication device during inhalation and exhalation of a user. In block 304, features are extracted from the audio data. In block 306, an indication of lung function is determined based, at least in part, on the features. In block 308, the indication of lung function is transmitted to the communication device.

[069] The system 214 of Figure 2 may be used to implement the method 300. For example, the executable instructions for measuring lung function 208 may include instructions for causing the computing system 202 to perform the actions in blocks 302, 304, 306, and 308. [070] In block 302, the audio data is received through a voice communication channel, such as the voice communication channel 218 of Figure 2. In some examples, the voice communication channel may compress the audio data. In some examples, such as when the voice communication channel is a GSM channel, the audio data may be compressed using linear predictive coding. Because the voice communication channel compresses, and in some examples also or instead encodes, the audio data, the methods used to determine an indication of lung function based on the audio data are advantageously sufficiently robust to operate on the compressed and/or encoded data on receipt tiirough the communication channel. For example, when the voice communication channel is implemented using a GSM channel or other communication channel which encodes voice using LPC, LPC features may advantageously be preserved through the channel and used to determine indications of lung function.

[071] Generally, GSM voice coding technologies use a source-filter model for speech, where representations of the filter created by the vocal tract and the source excitation created by the vocal cords are transmitted in lieu of the raw audio. The most common method for separating out the source excitation from the vocal tract filter is to use LPC. It isolates all linear relationships up to a certain order in the signal, which include the resonances due to the vocal tract's filtering. These resonances are also used in methods described herein for determining indications of lung function. As such, the LPC encoding generally preserves much of the information in the signal used to determine indications of lung function in accordance with examples described herein. The LPC features may be present in audio data received over a voice communication channel in some examples despite many smaller details being lost such as higher harmonics of the fundamental resonance and spectral energy above a threshold, such as 4 kHz. During transmission over a voice communication cha nel, the source excitation of the signal may be μ-Law encoded and transmitted. This encoding may compand the amplitude of the signal (e.g. compresses before transmission, expands after reception), causing a reduction in effective dynamic range at large signal values. Examples of methods described herein for determining indications of lung function may be sufficiently robust to operate in the presence of the reduction in effective dynamic range. [072] In some examples, low-energy components in the audio data may be suppressed by- transmission through the voice communication channel. For example, the energy of the signal may abruptly cut off in patches. Directly estimating FVC from the signal may therefore be more difficult than if low-energy signal components were not sacrificed. However measures such as FEV1 may still easily estimated, particularly in examples where the signal stays above a noise floor for the initial segment of a spirometry effort.

[073] In block 304, features may be extracted from the audio data, e.g. using the computing system 202 of Figure 2. The features extracted may generally fall in three categories: temporal envelope detection, spectrogram processing, and linear predictive coding (LPC). Accordingly, signal envelope features may be extracted, resonance tracking features may be extracted, linear predictive coding features may be extracted, or combinations thereof. In some examples, the voice communication channel may be implemented using a channel, such as a GSM channel, which uses LPC to encode voice. Accordingly, in some examples features extracted in block 304 based on LPC remain largely preserved.

[074] In block 306, an indication of lung function may be determined based, at least in part, on the features. Machine learning techniques may be employed to determine one or more indications of lung function based on the extracted features. For example, forced vital capacity (FVC), forced expiratory volume in one second (FEV1), peak expiratory flow (PEF), ratios of those, or combinations of tliose may be determined based, at least in part, on the features extracted in block 304. In some examples, other indications of lung function may be determined (e.g. pass/fail in accordance with a selected metric).

[075] Features may be weighted in performing a determination of lung function. The weighting may be dependent on the voice communication channel through which the audio data was received. For example, GSM communication channels may preserv e LPC features well, while other features may be more distorted by the communication channel. Accordingly, when the voice communication channel used to receive the audio data is a GSM channel (or other channel employing LPC encoding techniques), the LPC features extracted in block 304 may be weighted more heavily in a determination of lung function in block 306.

[076] In some examples, weights for the various features may be determined through a calibration procedure. In some examples, calibration spirometry efforts may be performed using a particular type (e.g. make, model) of communication device over a particular type (e.g. telephony network) of voice communication channel. Those efforts may be compared with spirometry- efforts performed on a clinical device. The comparison may be used to determine a set of weights for extracted features which are best able to predict the indication of lung function given by the clinical device performance. Weightings may be stored (e.g. accessible to the computing system of Figure 2) and used for audio data received from a same or similar communication device type over the same or similar voice communication channel. ] In some examples, one or more regressions may be performed on the features.

Examples of regressions include but are not limited to, linear regression, least-angle regression, elastic net regression, and kNN regression. In some examples, a linear regression, a least-angle regression, an elastic net regression, and a kNN regression may be performed. J A median of the regressions may be taken to determine one or more indications of lung function.

] In block 308, the indication of lung function may be transmitted to the communication device. The indication of lung function may be provided, for example, using audio or short messaging service. In some examples, the indication of lung function may be read back to a user of the communication device in a return call, or in a same call session as initiated to receive the audio data from the communication device.

J In some examples, an ensemble of machine learning algorithms (e.g. multiple regressions) may be used to process the extracted features into a predicted values in block 306, the median of which may be taken to be the prediction for that particular indication (e.g. spirometry measure). The machine learning algorithms generally may learn the relationship between the features and proper predicted values for the diagnostic measures; however these relationships can change depending on the phone model, cell earner and/or call quality used to provide the audio data from the user to the computing system used to perform the machine learning techniques.

J A dynamic "channel estimation" algorithm may send a predetermined signal (e.g. a prerecorded audio file of noise, or pilot tones played through the speakers of another device) over the voice communication channel (e.g. cell network) to probe it and determine relative weightings of the features being calculated. This channel estimation may allow the machine learning algorithm to adapt to microphones that heavily emphasize certain ranges of frequencies but not others and cell networks that employ more aggressive compression than their competitors.

[082] In some examples, the methods may include estimating channel state information of a voice communication channel. For example, the computing system 202 of Figure 2 may estimate channel state information of the voice communication channel 218. Channel state information generally refers to information regarding perturbations that may be imposed on a signal by the channel itself. Channel state information may, for example, refer to a frequency- response of the voice communication channel. In some examples, the executable instructions for measuring lung function 208 may include instructions for estimating channel state information (e.g. estimating the channel).

[083] The channel state information may be estimated, for example, by having a user of a communication device (or another transmitting device) transmit a known signal (e.g. a tone or sequence of tones) through the voice communication channel . In some examples, a pilot tone is swept across frequencies (e.g. 0kHz to 20kHz) at a constant amplitude. On receipt of signals received through the voice communication channel, a computing system (e.g. computing system 202 of Figure 2) may compare the received signals with the known signal and provide estimated channel state information based on the comparison. For example, in the case of the pilot tone swept across frequencies at a constant amplitude, the received amplitude may be measured, resulting in a measure of a frequency response of the voice communication channel . The frequency response may be used to tune (e.g. adjust) frequency-domain-based features extracted from the audio data.

[084] The channel state information may be used, for example, to select machine learning techniques and/or associations to be used in determining indication(s) of lung function from extracted features of the audio data.

[085] Figure 4 is a flowchart of a method arranged in accordance with examples described herein. The method 400 describes how one or more indications of lung function (e.g. Lung function estimate 420) may be determined based on audio data 402. The method 400 may be used to implement, for example, block 306 of Figure 3. The executable instructions for measuring lung function 208 may include instructions for performing the method 400 in some examples.

] Audio data 402 may be collected by a communication device, e.g. communication device 210, during a spirometry effort and transmitted over a voice communication channel. The audio data (e.g. audio samples) originally collected by the communication device may be a function of pressure because the microphone may be considered essentially an uncalibrated, AC-coupled pressure sensor. After compensating for pressure losses between the mouth and microphone, flow may be approximated from pressure and the AC-coupled artifacts reduced and/or removed.

] Sound is generally measured in pressure. Without being bound by theory, sound may also be generated during a subject's spirometry effort (e.g. forced expiratory maneuver) by the resulting expiratory airflow. For example, sound is generated by the expiratory airflow as it passes through the subject's vocal tract, through the subject's mouth, and through the environment surrounding the subject. In example sound-based spirometric devices, systems, and methods disclosed herein, audio data of sound generated during a subject's spirometry effort (e.g. forced expiratory maneuver) may be processed to generate expiratory flow-based pulmonary function data that may be used to assess the pulmonary function of the user.

] Features may be extracted from the audio data 402, The features may include, envelope features (e.g. Hiibert envelope 404), resonance tracking features (e.g. Resonance tracking 406), linear predictive coding features (e.g. Linear predictive coding 408), or combinations thereof. Examples of extraction of the features are described, for example, in co-pending U.S. Application Serial Number 14/400,064, filed May 10, 2013 entitled "Sound- based spirometric devices, systems, and methods," which application is hereby incorporated by reference in its entirety for any purpose, and U.S. Provisional Application No. 61/645,176, filed May 10, 2012, which application is incorporated herein by reference in its entirety for any purpose.

J The sound of a subject's spirometry effort (e.g. forced expiratory maneuver) at the microphone may be influenced by the flow rate of air from the subject's lungs and may include superfluous sounds generated by the expiratory airflow, for example, as the expiratory airflow passes through the subject's vocal tract, through the subject's mouth, and through the subject's surrounding environment. In many examples, one or more superfluous sounds generated by the expiratory airflow may be modeled and removed from, the audio data 402, thereby modifying the audio data 402 to remove or reduce pressure fluctuations that are less directly relatable to the rate of the spirometry effort (e.g. expiratory airflow). These modeled and removed superfluous sounds may be referred to herein as a first class of feature.

[090] The sound of a subject's spirometry effort (e.g. forced expiratory maneuver) at the microphone may also include additional sounds that, in many examples, may be used to infer expiratory flow rate because the intensity of these additional sounds may be related to the rate of airflow. Such additional sounds include items such as wind shear, vocal tract resonances, wheezes, and nasal resonances. In some examples, one or more of such additional sounds may be isolated and the intensity of the additional sound may be used as a feature.

[091] In some examples, sound pressure reductions that occur as the sound travels between the subject's mouth and the microphone and/or reverberation of sound in the subject's environment may be accounted for during processing of the audio data 402. For example, in some examples, inverse radiation modeling (e.g., a model of a spherical baffle in an infinite plane— also known as Flanagan's sound production model) may be used to account for the sound pressure reductions that occur as the sound travels between the subject's mouth and the microphone and/or reverberation of sound in the subject's environment. In some examples, the distance may be approximated from user's height and arm length, and may be adjusted by readings from an accelerometer (e.g., if a communication device including the microphone and the accelerometer is moved closer in).

[092] In some examples, the expiratory airflow rate (distance/time— e.g., meters/second) may be modeled from the combined features using non-parametric regression. In some examples, the volume of the expiratory airflow may be calculated by estimating the area of the subject's mouth opening and integrating the flow. For example, the flow rate of air can be measured in m/sec. The estimated area of the subject's mouth opening can be used to convert the flow rate of air to a volumetric flow rate (e.g., liters/second). The volumetric flow rate can be directly integrated to get volume.

[093] Audio data 402 may accordingly represent audio data as received from a communication device through a voice communication channel. The received data may be uncalibrated, AC-coupled measures of pressure, p(t), at the microphone of the communication device, and may have been modified by the voice communication channel (e.g. compressed and/or encoded). Tire audio data 402 may be processed to, for example, compensate for pressure losses as the sound travels from the user's mouth to the microphone, convert the pressure values to an approximation of flow, and remove and/or reduce the effects of AC coupling. In some examples, pressure losses may be approximated using an inverse model of the sound reverberation around the user's head. Turbulent airflow, as it passes through a fixed opening (e.g., the user's mouth), has a characteristic pressure drop, which, in many examples, may be used for converting pressure into flow. In some examples, at least one of: ( 1) signal power and frequency characteristics, and (2) models of the vocal tract may be used to remove and/or reduce effects of AC-coupling and refine the flow approximations.

[094] In some examples, regression may be used to combine these flow approximations and remove and/or reduce non-linearity. In some examples, the processing of the audio data 402 may include compensation and/or feature extraction and machine learning linear regression 410

[095] In some examples, a first stage in processing the audio data 402 may be to use inverse radiation modeling to compensate for pressure losses sustained over the distance from the user's mouth to microphone and for reverberation/reflections caused in and around the user's body. Any suitable inverse radiation modeling can be used. For example, the transfer function from the microphone to the user's mouth can be approximated by equation 1 , which is corresponds to a spherical baffle in an infinite plane.

[097] where: D_mm is the arm length (e.g., approximated from user's height); Chead is the head circumference (e.g., approximated from user's height); and c is the speed of sound. The transfer function inverse is applied by converting it to the time domain, hj_nv(t), and using Finite Impulse Response (FIR) filtering with the incoming audio data. Once applied, the output may be an approximation of the pressure at the lips, P'.j_ps(t). ] The pressure at the lips (pjj_ps(t)) may then be converted to a flow rate. For turbulent airflow, equation (2) is a non-linear equation that can be used to convert pressure drop across the lips to flow rate through the lips.

uii_Ps(t ~²⁷t^rupsJ²Pups(t) equation (2)

0] where: ru_ps is the radius of the user's mouth opening (e.g. a constant resistance across frequency) .

1] Some scaling constants have been removed from each of equation ( 1) and (2) and the equations are only proportional. The scaling constants were removed because p(t) is not calibrated, so uups(t) may only be proportional to the actual flow rate. Moreover, it is unclear how well these equations ( i) and (2) perform when using approximations of D_3rm, Chead, ari raps; and how non-linearity in the microphone affects inverse modeling. Therefore, in some examples, each measure p(t), pii_ps(t), and ui_sps(t) may be used for feature extraction and regression may be used to determine which features are most stable.

2] Each measure, p(t), pu_ps, (t), and uu_ps(t), may represent a high frequency, AC-coupled signal, from which a separate volumetric flow rate may be approximated. In some examples, approximating volumetric flow rate from, these signals includes using three transformations of these signals, referring again to Figure 4, these may include: (1) envelope detection, e.g. Hilbert envelope 404 (2) spectrogram processing, e.g. resonance tracking 406 and (3) linear predictive coding (LPC), e .g. linear predictive coding 408. Each envelope of the signals (p(t), Pii_ps, (t), and U_!i_ps(t)) can be assumed to be a reasonable approximation of the flow rate because it is a measure of the overall signal power (or amplitude) at low frequency. Spectrogram processing may be used to extract resonances. In the frequency domain, resonances may be assumed to be amplitudes excited by reflections in the vocal tract and mouth opening and therefore should be proportional to the flow rate that causes them . Linear prediction may then be used as a flow approximation. Linear prediction generally assumes that a signal can be divided into a source and a shaping filter and it estimates the source power and shaping filter coefficients. The "filter" in examples described herein may be an approximation of the vocal tract. The "source variance" may be an estimate of the white noise process exciting the vocal tract filter and is an approximation of the power of the flow rate from the lungs. [0103] Envelope features, such as Hilbert envelope 404, generally measure the energy of the audio data 402 over different frames. The low frequency envelope of the audio data (proportional to power here) may be extracted by squaring the signal and low pass filtering at a sub-1 Hz cutoff. The time domain envelope may also be taken using the Hilbert envelope. The Hilbert transform of the signal may be taken and added back to the original signal. Low pass filtering may then be used to extract the envelope using cascaded second order system filters. Each signal (p(t), pii_ps,(t), and uij_ps(t)) can be down-sampled 42 to have the same sampling rate as the spectrogram and linear prediction models. Different estimations of the envelopes can be ascertained from using slightly different low pass filters on the squared data and Hilbert transformed data. In one embodiment, 12, features are used.

ΘΙ 04] In some examples, during the spirometry effort (e.g. forced exhalation), the audio data 402 may be buffered, e.g. into 30 ms frames (with 50% overlap between frames). A spirometry exhalation typically lasts from four to seven seconds, resulting in 250-500 frames per exhalation. Each frame may then be windowed using a hamming window and the Fast Fourier Transform |FFT|<JB may be taken to produce the magnitude spectrogram of the signal . The resonances can be extracted (resonance tracking 406) using local maxima in each FFT frame, calculated over a sliding window, resulting in resonance tracking features. Any maxima that is greater than a suitable threshold, for example 20% of the global maximum, can be saved. After ail frames have been processed, in order to preserve only large and relatively long resonances, any resonance less than a threshold, e.g. 300 ms, can be discarded as noise. Finally, the average resonance magnitude in each frame may be calculated and saved as resonance tracking features.

[0105] In some examples, the audio data 402 may be again windowed into overlapping frames of, e.g. 30 ms. For each frame a number of LPC models can be taken (e.g. Linear predictive coding 408), for example, with filters orders of 2, 4, 8, 16 and 32 (increasing vocal tract complexity). The approximated "source power" that excites the filter can be saved for each frame as an approximation of the flow rate. The LPC may be taken from using p(t), Piip_s,(t), and/or uu_ps(t) . Various combinations of LPC order can be used to ascertain different estimates of the vocal tract. In one embodiment this results in 34 different estimates. The audio signal can be filtered by the inverse LPC model, which may leave only noise from random noises, such as wind shear. Additionally, the bandwidth and magnitude of the largest resonance from the LPC model can be calculated and used as another spectral estimate of the signal. By isolating the resonances between 400 and 900 Hz with different LPC models, the reverberation of the lips can be estimated as the lips typically resonate in this frequency range. In one embodiment, this adds to the number of spectral features, totaling 50.

] In some examples, once the approximated flow- rates are returned they may be denoised using a Savitsky-Golay polynomial filter, e.g. of order 3 and size 11. A third order polynomial can be fit inside a moving window and is robust to many types of noise while keeping the relative shape of the most prominent signal intact. The filtered and non-filtered signals may individually or both fed as features to the subsequent regression stage.

] Feature extraction, e.g. using Hilbert envelope 404, resonance tracking 406, and linear predictive coding 408, results in a number of uncalibrated approximations of the flow rate. These features may be used in a number of different regressions. In Figure 4, four regressions are used. The regressions may be carried out using, for example, the scikit-leam toolkit in Python and a leave-one-out cross-validation may be used in some examples to avoid overfitting.

] An example of four regressions is described with reference to Figure 4, however it is to be understood that the particulars of the regressions, including the number of regressions, may be different in other examples. In some examples, channel state information may be used to select a number, type, and/or parameters for regressions that are used to generate mdication(s) of lung function.

] One regression may be a linear regression 410 that tries to find a linear relationship between the features and groundtrath lung function value. A second regression may use a least-angle regression 412 (LARS) . LARS generally selects the most useful features using a variant of forward feature selection, but the underlying model is assumed linear. The third regression, elastic net regression 414, uses the elastic net algorithm, which eliminates features m a slightly different way than LARS. The elastic net regression 414 uses a combination of LASSO regression and ridge regression for regularizatson that is often more stable. Finally an encapsulated k-Nearest Neighbor regression (kNN regression 416) with (k = 2) is used. This regression finds the convex hull of the data in the feature space and fits a locally linear regression. Though the underlying model is assumed linear, the local fitting often can fit many different types of noniinearity. The final regression estimate may be found by taking the median (e.g. Median 418) of these four regressions. While four regressions are shown in Figure 4, other numbers of regressions may be used in other examples, including 1, 2, 3, 5, 6, or more regressions.

] The feature extraction, regressions, median, and/or lung function estimate stages shown in Figure 4 may be repeated for each desired indication of lung function (e.g. FEVl, FVC, and/or PEF indications). In some examples, certain features may be weighted more heavily in making a lung function estimate. For example, LPC features may be weighted more heavily when the audio data was received through a GSM or other voice communication channel employing LPC encoding techniques. In some examples, weights for the various features may be determined tlirough a calibration procedure, and used to adjust a combination of features used for estimating lung function.

] Figure 5 illustrates an example method arranged in accordance with examples described herein. The method 500 includes optionally dialing a call-in service at block 502, performing a spirometry effort at block 504, transmitting audio data collected by a communication device during the spirometry effort to a server through a voice communication channel in block 506, and receiving an indication of lung function corresponding to the audio data at the communication device in block 508.

] In block 502, a user may dial a call-in sendee. For example, a user may dial a toll-free or other phone number specifying a call-in sendee that may perform spirometry testing. The user may use a communication device (e.g. the communication device 210 of Figure 2) to call the call-in service. The dialed number may place a user's communication device in communication with a computing system (e.g. a remote server, e.g. computing system 202) configured to determine indications of lung function. The dialed number may place a user's communication device in communication with the computing system over a voice communication channel, and may be intermediated by other voice servers in some examples. Providing a call-in service may in some examples improve accessibility of spirometry testing - e.g. make spirometry testing available to anyone with access to a GSM cell phone network. [0113] In some examples, a user may be prompted to enter, using their communication device, certain biographical information. Biographical information that may be collected may include, but is not limited to, age, weight, gender, ethnicity, height, arm length, head circumference, health history, address, and combinations thereof.

[0114] In some examples, the communication device may play one or more known sounds (e.g. prerecorded audio and/or pushing selected buttons on a phone designed to generate predetermined tones). The known sounds may be provided to a server through the voice communication channel and may be used to estimate channel state information that may be used to guide the determination of one or more indication(s) of lung function as described herein.

[0115] In block 504, the user may perform a spirometry effort. Generally, the user may hold the communication device at arm's length and fully inhale and exhale. The full inhalation and exhalation may in some examples include a forced exhalation. Any suitable spirometry effort procedure may be used in some examples. The communication device may capture audio data during the spirometry effort, e.g. using a microphone of the communication device.

[0116] In block 506, the audio data collected by the communication device may be transmitted through the voice communication channel to a server. The server may determine one or more indications of lung function based on the audio data, in accordance with examples described herein, such as those described with reference to Figure 2- Figure 4.

[0117] In block 508, the user may receive, at their communication device, an indication of lung function. The indication of lung function can, for example, be texted back to the communication device (e.g. using SMS) or read aloud during the call.

[0118] Additionally or instead of receiving the indication of lung function in block 508, the communication device may receive feedback on the spirometry effort. For example, voice feedback may be provided if the amplitude of the audio data delivered by the spirometry effort was below a threshold, and generally not usable for the analysis described herein. In some examples, voice feedback may be provided that gave an indication to the user of how to improve the spirometry effort (e.g. breathe out fully, exhale longer, inhale more sharply, hold the phone closer). In some examples, audio coaching may be provided during the call (e.g. prior to and/or during the spirometry effort). The audio coaching may be provided by a remote coach and/or pre-recorded message and delivered to the communication device over the voice communication channel.

] Examples described herein may facilitate implementation of sound-based spirometry over a telephone network, which may extend the availability of spirometry testing.

Indications of lung function determined using systems and methods described herein may be utilized to treat conditions including, but not limited to, chronic obstructive pulmonary disease (COPD) and asthma.

] From the descriptions herein it will be appreciated that, although specific embodiments of the invention have been described herein for purposes of illustration, various modifications may be made without deviating from, the spirit and scope of the invention.

EXAMPLE

] We evaluated an example system in a controlled study with 50 patients. Each participant performed spirometry efforts with and without a whistle on two different phones recording the audio through the cell phone network and using a local application on two other smartphones. The participants used two different sizes of vortex whistles to determine whether different sizes work better for different demographics. The groundtruth was collected on two different FDA-approved spirometers. Our results show that the system, without a whistle, has a mean error of 6.2% for the four major clinically relevant lung function measures. For FEV1 %, the most commonly used measure, the mean error is 5.75%. With a whistle, the system has a mean error of 7.6% for the four measures, and 7% for FEV1%. Although, on average using the whistle leads to higher error in lung function estimates, it performs significantly better for people with lower lung function and produces fewer false negatives (over estimations of lung function). The system serves as a demonstration of how researchers can perform audio-based sensing on mobile phones by leveraging the voice channel .

] We created a dataset of audio samples and ground truth spirometry data. We recruited 50 participants (30 males), ranging in age from 21 to 67 years (M = 30) for 30-minute study sessions. 20% of participants had mild to moderate lung obstruction. [0123] The audio data was collected on four phones, two iPhone 4S smartphones, a Samsung Galaxy Note 3, and a Sony Ericsson W580i feature phone. All four phones were kept in front of the user at roughly an arm's length away. The distance was not formally controlled or varied. One of the iPhones and the Samsung Note were recording the audio data locally at 32 kHz and 44.1 kHz, respectively. The other two devices were sending the data over the GSM voice channel. These phones placed phone calls to different Google Voice accounts where the data was recorded in the form of voicemail messages. Google Voice saved the audio data in the form of MP3 files at 44.1 kHz, but the GSM channel band-limited the data to less than 8 kHz.

[0124] The data from the local phones (iPhone 4S and Samsung Note) was transferred to the computer over a USB connection at the end of the study. The data from the Google Voice accounts was downloaded as MP3 files to a computer.

[0125] The groundtruth for the participants was collected on two FDA-approved clinical spirometers, the nSpire Koko Legend and the NDD EasyWare spirometer. The two spirometers were used to answer two questions: (1) whether the participants got fatigued as the session progressed, and (2) how much variability exists between the output of the two devices, so we could use it as a benchmark for the system's performance. The participants performed at least 15 spirometry efforts (3 each for: 2 clinical spirometers, 2 whistles, and 1 without whistle). Spirometry measurements are completely effort-dependent and some fatigue can build up when performing this many efforts. Therefore, we recorded efforts on one clinical spirometer at the beginning of the session and on another spirometer at the end of the session. The order was randomized for each participant.

[0126] At the start of each session, the spirometry effort (e.g. forced expiratory maneuver) was explained to the participants and they were asked to practice using the spirometer. Once the participants were able to perform an acceptable maneuver according to the ATS criteria for reproducibility, tliree efforts were recorded using the spirometer. Next, the participants were introduced to the system. The study used a within-subjects 2x2x3 factorial design. The factors and levels were: [0127] · Phone Type: iPhone and non-iPhone. Two non-iPhone devices were used: Samsung Note and Sony Ericsson W580i. The W580i is a feature phone and was used to evaluate the performance of the system using an approximately 10-year-old device.

[0128] · Channel Type: Local recording or GSM. The iPhone was kept consistent in both channels to analyze the performance of the system if only the channel is changed.

[0129] · Whistle: No whistle, small whistle, and big whistle. We recorded audio data, for two whistles to understand if different participants preferred different sizes or if one size gave more reliable results that the other.

[0130] All the conditions were counterbalanced and the order of the whistles was randomized. The four phones (Phone Type χ Channel Type) recorded the audio simultaneously, thus saving the participants from performing tests with each device type separately. One of the system operators, trained to administer spirometry efforts, gave feedback to the participants regarding the acceptability and quality of the efforts. In the future, a system may automatically determine if an effort was too low in volume and the user has to retry.

[0131] Note that the system data and the clinical spirometer data was not collected at the same time so explicit ground truth was unknown. Instead, each effort from the system was associated with the best effort selected by the clinical spirometer. As per the ATS criteria, the spirometer selects the effort with the highest FVC as the best one.

[0132] Performance results of the system when compared to the two clinical spirometers is discussed in terms of accuracy of estimated lung function measures and false positives vs. false negatives. We break down these results by Phone Type and Channel Type. We also compare the performance of the sy stem with and without a whistle. Finally, we discuss the accuracy and usefulness of curves generated by the system. Based on our evaluation we conclude that the system can help in screening and monitoring patients with lung impairments in low resource regions.

[0133] As mentioned earlier, we used two clinical spirometers to collect the groundtruth. We compared their respective lung function measure and found that PEF had the maximum difference of 9.2% between the two devices, and FEV1, FVC, and FEV1 % had a difference of 5.1%, 5.2%, and 3.2%, respectively. However, none of these differences are statistically significant (based on an F-test, p>0.05). We also studied the effect of order to understand if fatigue played any role in exaggerating the difference between the two devices. In a 2 -way ANOV A test with presentation order of the two spirometers as a between-subjects factor, we found that the difference in estimates of PEF and FEVl were statistically significant (p<0.05). As mentioned earlier, spirometry is an effort dependent test, and as fatigue builds up, the peak rate at w hich patients can exhale (PEF) and how fast can they exhale (FEV l) tends to decrease. Our analysis shows that by the time the session ended many participants were fatigued, therefore we use the results from the first spirometer that the participants used as their groundtruth. We do not explicitly try to correct for fatigue in this example, however.

[0134] We break down the comparison of measurements from the system, and the groundtruth by evaluating how well it performs for different lung function measures and the number of outliers and false negatives.

[0135] For all lung function measures, the system, returns an average error of less than 10%.

There is no significant difference between the performance of smartphones processing the audio data locally (Samsung Note and Apple iPhone) and phones processing data received over the GSM channel - (Sony W580i and Apple iPhone 4S). The performance was best for FEVl %, which is the most common measure of lung function used in diagnosis because it is typically more consistent. The mean error rate for FEV1% is below 6% for all the four conditions. The ATS acceptability criteria require lung function measures to be within 7% to 10% of one another. This means that for most participants, the system performs well within the expected level of variation. However, it is important to evaluate the outliers and see whether the lung function measures are under-estimated or over-estimated.

[0136] The systems tended to over-estimate the actual value for patients with low lung function (FEV1 % < .75), e.g., a false negative. In any medical device, it is generally more acceptable to have a false positive than a false negative. The main reason the example system currently has more false negatives for low lung function is because the algorithm is data driven and the population with higher lung function is better represented. Therefore the model tends to bias towards the median value. Considering the signal-to-noise ratio is lower for the devices connected over the GSM channel, the false negatives are slightly more pronounced in case of systems analyzing the data received over the voice communication channel.

] One way to quantify the model's bias towards higher lung function is to calculate the statistical effect of lung function measure (FEV1% in this case) on the error of the model. We tested for effects of groundtrath FEV1% on the percent error through a chi-square test. We found that there was a significant effect of the groundtrath FEVl % on the accuracy of the system. As such, the performance of the example system might degrade further if tested on more highly obstructed patients. Although the bias is only slight and there are relatively few false negatives, from a diagnostic perspective, it could mean that patients are screened improperly.

Claims

CLAIMS What is claimed is:

1. A method comprising:

receiving, from a communication device, through a voice communication channel, audio data obtained by a microphone of the communication device during inhalation and exhalation of a user; and

transmitting, to the communication device, an indication of lung function of the user based on the audio data.

2. The method of claim 1 wherein the audio data is compressed by the voice communication channel.

3. The method of claim 2 wherein the audio data is compressed by the voice communication channel using linear predictive coding.

4. The method of claim 1 wherein the voice communication channel comprises a GSM channel.

5. The method of claim 1 wherein the indication of lung function is provided using audio or short messaging service.

6. The method of claim 1 further comprising extracting features from the audio data and determining the indication of lung function based, at least in part, on the features.

7. The method of claim 6, wherein the features comprise linear predictive coding features, signal envelope, and resonance tracking features.

8. The method of claim 6, wherein determining the indication of lung function comprises performing regression on the features.

9. The method of claim 8 wherein performing regression comprises performing a linear regression, a least-angle regression, an elastic net regression, and a kNN regression.

10. The method of claim 9 wherein determining the indication of lung function comprises taking a median of regressions performed.

1 1. The method of claim 6 further comprising estimating channel state information of the voice communication channel, and determining the indication of lung function based, at least in part, on the channel state information.

12. The method of claim 1 wherein the indication of lung function comprises forced vital capacity, forced expiratoiy volume in one second, peali expiratory flow, ratios of those, or combinations thereof.

13. A method for testing lung function, the method comprising:

holding a communication de vice at arm's length;

fully inhaling and exhaling;

transmitting audio data collected by the communication device during said fully inhaling and exhaling to a remote server through a voice communication channel; and

receiving an indication of lung function corresponding to the audio data at the communication device through audio or short messaging sen/ice communication.

14. The method of claim 13 wherein the voice communication channel comprises a GSM channel.

15. The method of claim 13 further comprising dialing, with the communication device, a predetermined phone number for a call-in service prior to said transmitting audio data.

16. The method of claim 13 wherein said fully inhaling and exhaling comprises performing a spirometry effort.

17. The method of claim 13 wherein said indication of lung function is based, at least m part, on features extracted from the audio data.

18. The method of claim 17 wherein said indication of lung function is based, at least in pari, on a linear regression, a least-angle regression, an elastic net regression, and a kNN regression of the features.

19. The method of claim 13, further comprising transmitting a set of pilot tones through the voice communication channel.

20. A system comprising:

a communication device; and

a computing system, the computing system including at least one processing unit and computer readable media encoded with instructions which, when executed by the at least one processing unit, cause the computing system to perform actions including;

receiving, from the communication device, through a voice communication channel, audio data obtained by a microphone of the communication device during inhalation and exhalation of a user; and

transmitting, to the communication device, an indication of lung function of the user based on the audio data.

21. The system of claim 20 wherein the computing system is a first computing system and wherein the communication device is configured to provide the audio data to a second computing system through the voice communication channel, and wherein the second computing system is configured to provide the audio data to the first computing system.

22. The system of claim 21 wherein the second computing system comprises a voice server of a telecommunications network.

23. The system of claim 20 wherein the indication of lung function is provided using audio or short messaging service.

24. The system of claim 20, wherein the actions further include extracting features from the audio data and determining the indication of lung function based, at least in part, on the features.

25. The system of claim 24, wherein actions further include performing regression on the features including performing a linear regression, a least-angle regression, an elastic net regression, and a kNN regression.

26. The system of claim 24, wherein the actions fuither include estimating channel state information of the voice comm unication channel, and determining the indication of l ung function based, at least in part, on the channel state information.