# US 2026/0061152 A1 · Sleep assistance through generation and evaluation of generative sleep content and/or sleep improvement interventions including training, adjusting, mediating, and/or integrating outputs of one or more ai models

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2026/0061152 A1
- Document ID: US20260061152A1
- Status: Published application
- Kind: A1
- Application number: 18/816,277
- Filed: 2024-08-27
- Published: 2026-03-05
- Earliest claimed: 2024-08-27
- Assignees: Stealth Labs Inc.
- Applicants: Stealth Labs Inc.
- Inventors: YADID AYZENBERG; AMIR LAZAROVICH; Michal Matuszczak
- Web page: https://patentgrep.com/patent/US20260061152A1

## CPC

- A61M 21/02: for inducing sleep or relaxation, e.g. by direct nerve stimulation, hypnosis, analgesia (for massage A61H; electrotherapy A61N, e.g. applying alternating or intermittent electric currents for producing anaesthesia A61N1/36021)
- A61B 5/0022: {Monitoring a patient using a global network, e.g. telephone networks, internet}
- A61B 5/02438: {with portable devices, e.g. worn by the patient}
- A61B 5/0816: {Measuring devices for examining respiratory frequency (measuring frequency of electric signals G01R23/00)}
- A61B 5/1115: {Monitoring leaving of a patient support, e.g. a bed or a wheelchair}
- A61B 5/1116: {Determining posture transitions}
- A61B 5/4812: {Detecting sleep stages or cycles}
- A61B 5/4815: {Sleep quality}
- A61B 5/4836: {Diagnosis combined with treatment in closed-loop systems or methods (A61B5/0036 takes precedence)}
- A61B 5/7267: {involving training the classification device}
- G10L 13/027: Concept to speech synthesisers; Generation of natural phrases from machine-based concepts (generation of parameters for speech synthesis out of text G10L13/08)
- G10L 15/22: Procedures used during a speech recognition process, e.g. man-machine dialogue
- G16H 20/70: relating to mental therapies, e.g. psychological therapy or autogenous training
- G16H 40/63: for local operation
- H04R 1/1016: {Earpieces of the intra-aural type}
- H04R 1/1041: {Mechanical or electronic switches, or control elements (switches in general H01H)}
- A61M 2021/0027: {by the hearing sense}
- A61M 2205/332: Force measuring means
- A61M 2205/3375: Acoustical, e.g. ultrasonic, measuring means
- A61M 2205/3553: remote, e.g. between patient's home and doctor's office
- A61M 2205/3592: using telemetric means, e.g. radio or optical transmission
- A61M 2205/52: with memories providing a history of measured variating parameters of apparatus or patient
- A61M 2205/80: voice-operated command
- A61M 2230/005: Parameter used as control input for the apparatus
- A61M 2230/06: Heartbeat rate only
- A61M 2230/42: Rate
- A61M 2230/63: Motion, e.g. physical activity
- G10L 13/00: Speech synthesis; Text to speech systems
- A61M: DEVICES FOR INTRODUCING MEDIA INTO, OR ONTO, THE BODY (introducing media into or onto the bodies of animals A61D7/00; means for inserting tampons A61F13/26; devices for administering food or medicines orally A61J; containers for collecting, storing or administering blood or medical fluids A61J1/05); DEVICES FOR TRANSDUCING BODY MEDIA OR FOR TAKING MEDIA FROM THE BODY (surgery A61B; chemical aspects of surgical articles A61L); DEVICES FOR PRODUCING OR ENDING SLEEP OR STUPOR {(Electrotherapy, e.g. producing anaesthesia by the use of alternating or intermittent currents A61N1/36021)}
- A61B: DIAGNOSIS; SURGERY; IDENTIFICATION (usefulness limited to only animals A61D)
- G10L: SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G16H: HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- H04R: LOUDSPEAKERS, MICROPHONES, GRAMOPHONE PICK-UPS OR LIKE ACOUSTIC ELECTROMECHANICAL TRANSDUCERS; ELECTRIC HEARING AIDS; PUBLIC ADDRESS SYSTEMS (producing sounds with frequency not determined by supply frequency G10K)

## Abstract

Disclosed are a system, a device, and/or a method of sleep assistance through generation and evaluation of generative sleep content and/or sleep improvement interventions including training, adjusting, mediating, and/or integrating outputs of one or more AI models. In one embodiment, a system includes an earbud generating an audio generation request through a voice interface, which is parsed by a sleep assistance server to extract a narrative prompt. A generative content server inputs the narrative prompt into an artificial neural network to generate a text data, and then text-to-speech model to generate a narrative audio. A content integration routine may overlay the narrative audio data with additional music and/or physiological guidance data. A generative audio data is returned to the earbud to assist in achieving sleep. The system may evaluate physiological data to determine effectiveness of the sleep assistance audio, and may augment, fine-tune, and/or retrain one or more AI models.

## Claims

### Claim 1

A system for generating customized audio for increasing effectiveness of relaxation and/or sleep, the system comprising:
a network;
an earbud wearable by a user comprising:
a speaker,
a microphone,
a processor of the earbud,
a memory of the earbud that is a non-transient computer readable memory of the earbud,
a wireless network interface controller communicatively coupled to the network, and
a request generation routine comprising computer readable instructions that when executed on the processor of the earbud generate an audio generation request through a voice interface implemented on the speaker and the microphone;
a sleep assistance server communicatively coupled to the network, comprising:
a processor of the sleep assistance server,
a memory of the sleep assistance server that is a non-transient computer readable memory of the sleep assistance server,
a request agent comprising computer readable instructions that when executed receive the audio generation request,
a content prompt extraction routine comprising computer readable instructions that when executed:
parse the audio generation request to extract a narrative prompt, and
store the narrative prompt in the memory of the sleep assistance server; and
a generative content server communicatively coupled to the network, comprising:
a processor of the generative content server,
a memory of the generative content server,
a text generation routine comprising computer readable instructions that when executed:
input the narrative prompt into a text generation model comprising an artificial neural network of the text generation model,
wherein the artificial neural network comprising a plurality of nodes comprising a set of input nodes, a set of hidden nodes, and a set of output nodes, and
generate a text data as the output of the artificial neural network of the text generation model,
a voice generation routine comprising computer readable instructions that when executed:
input the text data into a text-to-speech model, and
store a narrative audio data as an output of the text-to-speech model,
a content integration routine comprising computer readable instructions that when executed generate a generative audio data comprising an overlay of the narrative audio data and at least one of a music audio data, a physiological guidance data, and an ambient audio data, and
a generative content engine comprising computer readable instructions that when executed transmit the generative audio data to the earbud of the user to assist the user in achieving at least one of relaxation and sleep through the customized audio.

### Claim 2 (depends on 1)

The system of claim 1,
wherein the earbud further comprising:
an inertial measurement unit, and
a physiological signal agent comprising computer readable instructions that when executed gather a physiological data of the user from the inertial measurement unit of the earbud worn by the user while the generative audio data plays sound on the speaker of the earbud,
wherein the physiological data comprising at least one of motion of the user comprising a heartbeat, a respiration, and a macro movement; and
wherein the sleep assistance server further comprising:
a state monitoring routine comprising computer readable instructions that when executed utilize the physiological data gathered from the inertial measurement unit to determine the user is in at least one of a sleep state and an awake state over a time period,
a sleep metric routine comprising computer readable instructions that when executed determine one or more sleep metrics for a sleep session comprising a sleep onset latency value, a length of a sleep period during the time period, a ratio of the sleep period to an awake period during the time period, a number of REM periods, a length of REM periods, a number of Non-REM periods, a length of Non-REM periods, a number of interstitial awake periods, and a length of the awake period during the time period, and
a sleep content evaluation routine comprising computer readable instructions that when executed:
extract one or more generative elements of at least one of the narrative audio data and the text data associated with the narrative audio data, and
store one or more generative elements in association with an effectiveness value and an effectiveness rating based on the one or more sleep metrics.

### Claim 3 (depends on 1)

The system of claim 1, wherein the generative content server further comprising:
a general augmentation routine comprising computer readable instructions that when executed:
querying a general augment data comprising at least one of general augment narrative data, general augment music data, general augment ambient data, and general augment voice data,
extract a subset of the general augment data based on textual association with the audio generation request, and
load the subset of the general augment data into at least one of an input prompt of an artificial neural network and a context window of the artificial neural network,
wherein the input prompt comprises at least one of the narrative prompt, a music prompt, and an ambient prompt, and
wherein the artificial neural network is at least one of an artificial neural network of the text generation model.

### Claim 4 (depends on 1)

The system of claim 1, further comprising:
a specific augmentation routine comprising computer readable instructions that when executed:
query a user specific augment data comprising at least one of user augment narrative data, user augment music data, user augment ambient data, and user augment voice data,
extract a subset of the user specific augment data relevant to the audio generation request,
overwrite at least some of the subset of the general augment data within at least one of the input prompt of the artificial neural network and a context window of the artificial neural network, and
load the subset of the user specific augment data into at least one of the input prompt of the artificial neural network and the context window of the artificial neural network.

### Claim 5 (depends on 1)

The system of claim 1,
wherein the generative content server further comprising:
a model training routine comprising computer readable instructions that when executed retrain the artificial neural network of the text generation model based on at least one of the effectiveness value and the effectiveness rating,
wherein retraining comprises adjusting a parameter of at least one of the artificial neural network of the text generation model,
wherein adjusting the parameter comprising modifying a node weight of an ANN node,
wherein tuning the parameter adjusts a weight value of at least one node of the set of input nodes, the set of hidden nodes, the set of output nodes,
wherein the text generation model is a large language model, and
wherein the text data input into a voice synthesizer comprising a custom voice model; and
a guidance prioritization subroutine comprising computer readable instructions that when executed constrain the output of the artificial neural network of a narrative generation model to produce the text data in which a text clause of the text data is temporally associable with a physiological guidance element of a physiological guidance template,
wherein the text-to-speech model generates the narrative audio such that a voiceover of the text data is temporally associated with a physiological guidance element.

### Claim 6

A method for assisting a user to increase sleep effectiveness, the method comprising:
gathering physiological data from a user from a sensor of an earbud;
initiating a sleep session data;
monitoring a cognitive state of the user for a time period,
wherein the cognitive state comprising a sleep state determined based on at least one of a respiration rate, a respiration rate variability, a respiration rate curve, a heartrate, a heartrate variability, and a heartrate curve;
determining the user is in a sleep state based on the physiological data;
determining the user is in a concerned awake state based on the physiological data;
calculating one or more sleep metrics comprising a sleep onset latency value, a length of a sleep period during the time period, a ratio of the sleep period to an awake period during the time period, a number of REM periods, a length of REM periods, a number of Non-REM periods, a length of Non-REM periods, a number of interstitial awake periods, and a length of the awake period during the time period; and
reporting the one or more sleep metrics to the user through a voice interface using a speaker of the earbud upon determination the user is in the concerned awake state to assist the user in evaluating the sleep effectiveness.

### Claim 7 (depends on 6)

The method of claim 6, further comprising:
receiving a first user request for a sleep assistance audio;
initiating the sleep assistance audio generated from a first audio data on an earbud worn by a user,
determining an effectiveness value of the sleep assistance audio utilizing at least one of the sleep metrics of the time period compared against at least one of a general sleep metric baseline and a user baseline sleep metric generated with data comprising one or more previous time periods;
querying the user for an effectiveness rating of the sleep assistance audio over at least one of the time period and a segment of the time period;
receiving the effectiveness rating from the user; and
generating a sleep audio effectiveness data by associating (i) at least one of the effectiveness value and the effectiveness rating and (ii) at least one of an identifier of the first audio data and an element of the first audio data; and
storing the sleep audio effectiveness data association with a user profile of the user.

### Claim 8 (depends on 6)

The method of claim 6, further comprising:
inputting into an artificial neural network at least one of (i) the sleep data, and (ii) the one or more sleep metrics,
wherein the artificial neural network at least one of trained with and fine-tuned with data comprising one or more sleep improvement interventions associated with at least one of sleep data and a set of sleep metrics;
outputting a first sleep improvement intervention of the one or more sleep improvement interventions;
determining the user is initiating the sleep session; and
reporting the first sleep improvement intervention to the user through a voice interface using a speaker of the earbud,
wherein the first sleep improvement intervention comprising at least one of a decreased environmental noise, an increased sleep period, an earlier sleep time, a later awakening time, and utilization of a different audio data to generate a sleep assistance audio.

### Claim 9 (depends on 7)

The method of claim 7, further comprising:
receiving a second user request for the sleep assistance audio;
determining that at least one of that the effectiveness value is below a threshold value and the effectiveness rating is below a threshold rating; and
initiating the sleep assistance audio generated from a second audio data on the earbud worn by the user.

### Claim 10 (depends on 7)

The method of claim 7, the method further comprising:
receiving a second user request for the sleep assistance audio;
parsing the second user request to determine a requested element of the sleep assistance audio of the second user request;
generating a prompt from the second user request for at least one of selection of the sleep assistance audio and generation of the sleep assistance audio;
initiating a retrieval augmented generation data, referred to as a session RAG, for use in association with the sleep session;
querying the user profile and loading a description of the first audio data and at least one of the effectiveness rating and the effectiveness value into the session RAG; and
submitting the prompt and the session RAG to a large language model comprising an artificial neural network.

### Claim 11 (depends on 8)

The method of claim 8, further comprising:
receiving a sleep improvement request from the user,
wherein the sleep improvement request received through a voice interface of an earbud worn by the user;
generating a prompt comprising data extracted from the sleep improvement request;
submitting the prompt to a large language model comprising an artificial neural network of a sleep assistance model,
wherein the artificial neural network of the sleep assistance model at least one of trained with and fine-tuned with a training data comprising a text question related to sleep effectiveness paired with a text answer related to sleep effectiveness.
receiving an output from the artificial neural network of the sleep assistance model comprising a predictive text of the sleep assistance model; and
providing the predictive text of the sleep assistance model to the user through the voice interface as a second sleep intervention.

### Claim 12 (depends on 11)

The method of claim 11, further comprising:
reporting the effectiveness value to the user upon determination the user is in the concerned awake state;
generating a summary text data of the one or more sleep metrics;
inputting the summary text into a text-to-speech model;
generating a summary audio; and
transmitting the summary audio to the earbud worn by the user,
wherein the sleep assistance audio generated from at least one of a music audio data, a physiological guidance template, a narrative audio data, and a ambient sound data,
wherein the sleep assistance audio comprises a generative audio that is output from an artificial neural network of a text generation model, an artificial neural network of a music generation model, and an artificial neural network of a ambient sound generation model, and
wherein the sleep session persists until an automatic determination of a concerted awake state.

### Claim 13

A method for generating customized audio for increasing effectiveness of relaxation and/or sleep, the method comprising:
receiving an audio generation request through a voice interface collected on a microphone of an earbud;
parsing the audio generation request to extract a narrative prompt and a physiological guidance prompt;
storing the narrative prompt in a computer readable memory;
determining a narrative modifier from the audio generation request comprising at least one of a narrative style description and a narrative genre description;
inputting the narrative prompt into a text generation model comprising an artificial neural network of the text generation model;
generating a text data as an output of the artificial neural network of the text generation model;
inputting the text data into a text-to-speech model;
outputting a narrative audio data;
storing the physiological guidance prompt in the computer readable memory;
determining a physiological guidance modifier from the audio generation request comprising a physiological guidance type,
wherein the physiological guidance type comprising at least one of a respiration rate, a respiration pattern, and a heart rate;
inputting the physiological guidance prompt into a physiological guidance model comprising an artificial neural network of the physiological guidance generation model,
generating a physiological guidance template as an output of the artificial neural network of the physiological guidance model;
wherein the physiological guidance comprising one or more physiological
generating a generative audio data comprising an overlay of the narrative audio data and audio generated from the physiological guidance template; and
transmitting the generative audio data to the earbud of a user to assist the user in achieving at least one of relaxation and sleep through the customized audio.

### Claim 14 (depends on 13)

The method of claim 13, further comprising:
parsing the audio generation request to further determine if a music prompt is included within the audio generation request;
generating, if the music prompt was not present when the audio generation request was parsed, the music prompt by inputting the narrative prompt into a text-music relation model relating a text to at least one of musical elements, a music style description, and a music genre description;
optionally extracting a music modifier comprising at least one of a music filter, the music style description, and the music genre description;
storing in the computer readable memory the music prompt and optionally at least one of the music style description, the music genre description, and the music filter;
inputting the music prompt into a music generation model comprising an artificial neural network of the music generation model,
wherein the music generation model trained with training data comprising associations between text tokens and musical elements; and
generating a music audio data as an output of the artificial neural network of the music generation model,
wherein the generative audio data further comprising an overlay of the music audio data.

### Claim 15 (depends on 14)

The method of claim 14, further comprising:
parsing the audio generation request to further determine an ambient sound prompt;
storing the ambient sound prompt in the computer readable memory;
receiving an ambient modifier comprising at least one of an ambient filter and an ambient style description;
inputting the ambient sound prompt and the ambient modifier into an ambient sound generation model comprising an artificial neural network of the ambient sound generation model,
wherein the ambient sound generation model trained with training data comprising associations between text tokens and sound elements; and
generating an ambient audio data as an output of the artificial neural network of the ambient generation model,
wherein the generative audio data further comprising an overlay of the ambient audio data.

### Claim 16 (depends on 15)

The method of claim 15, further comprising:
querying a general augment data comprising at least one of general augment narrative data, general augment music data, general augment ambient data, and general augment voice data;
extracting a subset of the general augment data based on textual association with the audio generation request; and
loading the subset of the general augment data into at least one of an input prompt of an artificial neural network and a context window of the artificial neural network,
wherein the input prompt comprises at least one of the narrative prompt, the music prompt, and the ambient prompt, and
wherein the artificial neural network is at least one of an artificial neural network of the text generation model, an artificial neural network of the music generation model, and an artificial neural network of the ambient sound generation model.

### Claim 17 (depends on 16)

The method of claim 16, further comprising:
querying a user specific augment data comprising at least one of user augment narrative data, user augment music data, user augment ambient data, and user augment voice data,
extracting a subset of the user specific augment data relevant to the audio generation request;
overwriting at least some of the subset of the general augment data within at least one of the input prompt of the artificial neural network and the context window of the artificial neural network; and
loading the subset of the user specific augment data into at least one of the input prompt of the artificial neural network and the context window of the artificial neural network.

### Claim 18 (depends on 17)

The method of claim 17, further comprising:
constraining the output of the artificial neural network of the ambient sound generation model to produce an ambient audio data in which an ambient element is temporally associated with a physiological guidance element;
constraining the output of the artificial neural network of the music generation model to produce the music audio data in which a musical element is temporally associated with a physiological guidance element; and
constraining the output of the artificial neural network of the text generation model to produce the text data in which a text clause of the text data is temporally associable with a physiological guidance element,
wherein the text-to-speech model generates the narrative audio such that a voiceover of the text data is temporally associated with a physiological guidance element.

### Claim 19 (depends on 18)

The method of claim 18, further comprising:
gathering a physiological data of the user from a sensor of an earbud worn by the user while the generative audio data plays sound on a speaker of the earbud;
determining the user is in a sleep state;
determining one or more sleep metrics comprising a sleep onset latency value, a length of a sleep period during a time period, a ratio of the sleep period to an awake period during the time period, a number of REM periods, a length of REM periods, a number of Non-REM periods, a length of Non-REM periods, a number of interstitial awake periods, and a length of the awake period during the time period;
extracting one or more generative elements of at least one of the generative audio data, the narrative audio data, the text data associated with the narrative audio data, the music audio data, the ambient audio data, and the one or more physiological guidance elements;
storing the one or more generative elements in association with an effectiveness value and an effectiveness rating; and
retraining at least one of the artificial neural network of the text generation model, the artificial neural network of the music generation model, the artificial neural network of the physiological guidance model and the artificial neural network of the ambient generation model,
wherein retraining comprises adjusting a parameter of at least one of the artificial neural network of the text generation model, the artificial neural network of the music generation model, and the artificial neural network of the physiological guidance model and the artificial neural network of the ambient generation model, and
wherein adjusting the parameter comprising modifying a node weight of an ANN node.

### Claim 20 (depends on 19)

The method of claim 19, further comprising:
querying a prerecorded audio comprising at least one of a recorded narrative audio, a recorded music audio, a recorded ambient audio, and a recorded physiological guidance audio, and
integrating the prerecorded audio with the generative audio,
wherein the artificial neural network each comprising a plurality nodes comprising a set of input nodes, a set of hidden nodes, and a set of output nodes,
wherein tuning the parameter adjusts a weight value of at least one node of the set of input nodes, the set of hidden nodes, and the set of output nodes,
wherein the text generation model is a large language model, and
wherein the text data input into a voice synthesizer comprising a custom voice model.