# US 12,004,851 B1 · Methods for automatic cough detection and uses thereof > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 12,004,851 B1 - Document ID: US12004851B1 - Status: Granted - Kind: B1 - Application number: 18/429,130 - Filed: 2024-01-31 - Granted: 2024-06-11 - Earliest claimed: 2023-01-31 - Assignees: HYFE Inc. - Applicants: Iulian-Alexandru Circo; Joseph Russell Brew; Paul Simon Rieger; Peter McMichael Small; Georgios Kafentzis - Inventors: Iulian-Alexandru Circo; Joseph Russell Brew; Paul Simon Rieger; Peter McMichael Small; Georgios Kafentzis - Web page: https://patentgrep.com/patent/US12004851B1 ## CPC - A61B 5/0823: {Detecting or evaluating cough events} - A61B 5/7264: {Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems} - A61B 5/7267: {involving training the classification device} - A61B 5/7282: {Event detection, e.g. detecting unique waveforms indicative of a medical condition (cough events A61B5/0823; seizures A61B5/4094; sleep apnoea A61B5/4818)} - A61B 7/003: {Detecting lung or respiration noise} - A61B: DIAGNOSIS; SURGERY; IDENTIFICATION (usefulness limited to only animals A61D) ## Abstract A method of automatically detecting cough events from a continuous recording of ambient sound includes a step of detecting an onset of a possible cough event by analyzing acoustic energy distribution of an audio snippet in a frequency-time domain at a frequency above 100 Hz and classifying the event as a cough or a non-cough using convolutional neural networks. The method may be used for improved diagnosis in differentiating between various diseases having cough as a common symptom, monitoring the progression of a disease or efficacy of a treatment and for other purposes. ## Claims ### Claim 1 A method of automatically detecting cough events comprising the following steps: a. continuously recording ambient sound, b. upon detecting a change in ambient sound exceeding a predefined first threshold, identifying an onset of a possible cough event, c. recording an audio snippet including the onset of a possible cough event and continuing for a predefined audio snippet duration exceeding 100 msec thereafter, d. classifying the audio snippet recorded in step (c) as cough or non-cough based on analyzing acoustic energy distribution of the audio snippet in a frequency-time domain at a frequency above 100 Hz, e. discarding all identified non-cough events, f. repeating steps (b) through (e) if further possible cough events are identified after the audio snippet recorded in step (c), and g. compiling a record of all cough events detected during the recording of ambient sound in step (a). ### Claim 2 (depends on 1) The method of automatically detecting cough events, as in claim 1, wherein in step (g) the record of all cough events comprises none or at least one time stamp associated with the cough event classified in step (d). ### Claim 3 (depends on 1) The method of automatically detecting cough events, as in claim 1, wherein in step (d) classifying the audio snippet recorded in step (c) as cough or non-cough further comprises a step of comparing the acoustic energy distribution of the audio snippet against a predefined second threshold. ### Claim 4 (depends on 1) The method of automatically detecting cough events, as in claim 1, wherein in step (d) classifying the audio snippet recorded in step (c) as cough or non-cough comprises a step of using a statistical classifier. ### Claim 5 (depends on 4) The method of automatically detecting cough events, as in claim 4, wherein the statistical classifier is a neural network classifier. ### Claim 6 (depends on 5) The method of automatically detecting cough events, as in claim 5, wherein the neural network classifier is a convolutional neural network classifier. ### Claim 7 (depends on 6) The method of automatically detecting cough events, as in claim 6, wherein the convolutional neural network is pre-trained on a plurality of prior recordings of known human cough and non-cough events. ### Claim 8 (depends on 7) The method of automatically detecting cough events, as in claim 7, wherein the plurality of prior recordings of known human cough and non-cough events are processed to define a probability measure of the event to be a cough event or a non-cough event. ### Claim 9 (depends on 8) The method of automatically detecting cough events, as in claim 8, wherein the plurality of prior recordings of known human cough and non-cough events are processed by analyzing acoustic energy distributions of each recording in the frequency-time domain. ### Claim 10 (depends on 1) The method of automatically detecting cough events, as in claim 1, wherein in step (c) the predefined audio snippet duration does not exceed 1 sec after the onset of the possible cough event. ### Claim 11 (depends on 1) The method of automatically detecting cough events, as in claim 1, wherein in step (c) recording the audio snippet starts prior to the onset of the possible cough event. ### Claim 12 (depends on 11) The method of automatically detecting cough events, as in claim 11, wherein in step (c) recording the audio snippet starts at least 20 msec prior to the onset of the possible cough event. ### Claim 13 (depends on 1) The method of automatically detecting cough events, as in claim 1, wherein in step (b) detecting a change in acoustic energy comprises a step of detecting individual changes in acoustic energy in a plurality of bins, wherein each bin corresponds to a predefined range of acoustic frequencies. ### Claim 14 (depends on 13) The method of automatically detecting cough events, as in claim 13, wherein a predefined second threshold corresponds to a number of bins from the plurality of bins in which a detected change in acoustic energy exceeds a predefined acoustic energy limit. ### Claim 15 (depends on 1) The method of automatically detecting cough events, as in claim 1, wherein step (a) further comprises a step of removing silence sounds identified using a predefined silence threshold. ### Claim 16 (depends on 1) The method of automatically detecting cough events, as in claim 1, wherein in step (d), a step of analyzing the audio snippet comprises a step of dividing the audio snippet into a plurality of time frames. ### Claim 17 (depends on 16) The method of automatically detecting cough events, as in claim 16, wherein in step (d), all of the time frames of the plurality of time frames overlap with adjacent time frames.