# US 11,308,978 B2 · Systems and methods for energy efficient and low power distributed automatic speech recognition on wearable devices

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

- Number: US 11,308,978 B2
- Document ID: US11308978B2
- Status: Granted
- Kind: B2
- Application number: 16/531,500
- Filed: 2019-08-05
- Published: 2019-11-21
- Granted: 2022-04-19
- Earliest claimed: 2017-03-31
- Assignees: Intel Corporation
- Applicants: Intel Corporation
- Inventors: Binuraj K. Ravindran; Francis M. Tharappel; Prabhakar R. Datta; Tobias Bocklet; Maciej Muchlinski; Tomasz Dorau; Josef G. Bauer; Saurin Shah; Georg Stemmer
- Web page: https://patentgrep.com/patent/US11308978B2

## CPC

- G10L 25/24: the extracted parameters being the cepstrum
- G10L 15/08: Speech classification or search
- G10L 15/30: Distributed recognition, e.g. in client-server systems, for mobile phones or network applications
- G10L 15/02: Feature extraction for speech recognition; Selection of recognition unit
- G10L 19/032: Quantisation or dequantisation of spectral components
- G10L 25/78: Detection of presence or absence of voice signals (switching of direction of transmission by voice frequency in two-way loud-speaking telephone systems H04M9/10)
- G10L 2015/088: {Word spotting}
- H04W 4/80: Services using short range communication, e.g. near-field communication [NFC], radio-frequency identification [RFID] or low energy communication
- G10L: SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- H04W: WIRELESS COMMUNICATION NETWORKS (broadcast communication H04H; communication systems using wireless links for non-selective communication, e.g. wireless extensions H04M1/72)

## Abstract

Methods, apparatus, systems and articles of manufacture are disclosed for distributed automatic speech recognition. An example apparatus includes a detector to process an input audio signal and identify a portion of the input audio signal including a sound to be evaluated, the sound to be evaluated organized into a plurality of audio features representing the sound. The example apparatus includes a quantizer to process the audio features using a quantization process to reduce the audio features to generate a reduced set of audio features for transmission. The example apparatus includes a transmitter to transmit the reduced set of audio features over a low-energy communication channel for processing.

## Claims

### Claim 1

A wearable device comprising:
memory to store instructions;
processor circuitry to execute the instructions to:
identify an input audio signal including a sound to be evaluated;
identify one or more keywords in the sound;
when the one or more keywords are identified, extract a plurality of audio features from the sound, the plurality of audio features represented by a plurality of Mel-frequency cepstral coefficients (MFCCs); and
execute a machine learning quantization model to process the plurality of MFCCs to reduce a bit width of the plurality of audio features to generate a reduced set of audio features, the machine learning quantization model to implement a quantization selected based on an operating condition; and
communication circuitry to:
transmit the reduced set of audio features to a cloud computing device in response to the identification of the one or more keywords; and
obtain a response to the reduced set of audio features from the cloud computing device to cause an output.

### Claim 2 (depends on 1)

The wearable device of claim 1, further including a battery and a microphone.

### Claim 3 (depends on 1)

The wearable device of claim 1, wherein the bit width reduction is to reduce the bit width to eight bits.

### Claim 4 (depends on 1)

The wearable device of claim 1, wherein the bit width reduction includes a quantization of the plurality of audio features.

### Claim 5 (depends on 1)

The wearable device of claim 1, wherein the bit width reduction includes a compression of the plurality of audio features.

### Claim 6 (depends on 1)

The wearable device of claim 1, wherein the computing device includes a cloud-based processing system.

### Claim 7 (depends on 1)

The wearable device of claim 1, wherein the output includes a command to be executed by at least one of the wearable device or the computing device.

### Claim 8 (depends on 1)

The wearable device of claim 1, wherein the output includes feedback to the wearable device to adjust at least one of the extraction of the plurality of audio features or the reduction in bit width.

### Claim 9

At least one computer-readable storage medium comprising instructions that, when executed, cause at least one processor of a wearable device to:
obtain an input audio signal including a sound to be evaluated;
identify, when triggered by the input audio signal, one or more keywords in the sound;
extract, when the one or more keywords is identified, a plurality of audio features from the sound, the plurality of audio features represented by a plurality of Mel-frequency cepstral coefficients (MFCCs);
execute a machine learning quantization model to process the plurality of MFCCs to reduce a bit width of the plurality of audio features to generate a reduced set of audio features, the machine learning quantization model to implement a quantization selected based on an operating condition;
transmit the reduced set of audio features to a cloud computing device in response to the identification of the one or more keywords; and
obtain a response to the reduced set of audio features from the cloud computing device to cause an output.

### Claim 10 (depends on 9)

The at least one computer-readable storage medium of claim 9, wherein the instructions, when executed, cause the at least one processor to reduce the bit width to eight bits.

### Claim 11 (depends on 9)

The at least one computer-readable storage medium of claim 9, wherein the instructions, when executed, cause the at least one processor to reduce the bit width using a quantization of the plurality of audio features.

### Claim 12 (depends on 9)

The at least one computer-readable storage medium of claim 9, wherein the instructions, when executed, cause the at least one processor to reduce the bit width using a compression of the plurality of audio features.

### Claim 13 (depends on 9)

The at least one computer-readable storage medium of claim 9, wherein the at least one computer-readable storage medium is included in the wearable device, and wherein the instructions, when executed, cause the at least one processor to obtain the response to cause the output including a command to be executed by at least one of the wearable device or the computing device.

### Claim 14 (depends on 13)

The at least one computer-readable storage medium of claim 13, wherein the instructions, when executed, cause the at least one processor to obtain the response to cause the output including feedback to the wearable device to adjust at least one of the extraction of the plurality of audio features or the reduction in bit width.

### Claim 15

A method comprising:
identifying, by executing an instruction using at least one processor of a wearable device, an input audio signal including a sound to be evaluated;
identifying, by executing an instruction using the at least one processor, one or more keywords in the sound;
when the one or more keywords are identified, extracting, by executing an instruction using the at least one processor, a plurality of audio features from the sound, the plurality of audio features represented by a plurality of Mel-frequency cepstral coefficients (MFCCs); and
executing, by executing an instruction using the at least one processor, a machine learning quantization model to process the plurality of MFCCs to reduce a bit width of the plurality of audio features to generate a reduced set of audio features, the machine learning quantization model to implement a quantization selected based on an operating condition;
transmitting, by executing an instruction using the at least one processor, the reduced set of audio features to a cloud computing device in response to the identification of the one or more keywords; and
obtaining, by executing an instruction using at least one processor, a response to the reduced set of audio features from the cloud computing device to cause an output.

### Claim 16 (depends on 15)

The method of claim 15, wherein reducing the bit width includes at least one of: a) reducing the bit width using a quantization of the plurality of audio features, or b) reducing the bit width using a compression of the plurality of audio features.

### Claim 17 (depends on 15)

The method of claim 15, wherein obtaining the response to cause the output includes at least one of: a) obtaining the response to cause the output including a command to be executed by at least one of the wearable device or the computing device, or b) obtaining the response to cause the output including feedback to the wearable device to adjust at least one of the extraction of the plurality of audio features or the reduction in bit width.

### Claim 18

A wearable apparatus comprising:
means for processing to at least:
identify an input audio signal including a sound to be evaluated;
identify one or more keywords in the sound;
when the one or more keywords are identified, extract a plurality of audio features from the sound, the plurality of audio features represented by a plurality of Mel-frequency cepstral coefficients (MFCCs); and
execute a machine learning quantization model to process the plurality of MFCCs to reduce a bit width of the plurality of audio features to generate a reduced set of audio features, the machine learning quantization model to implement a quantization selected based on an operating condition; and
means for communicating to at least:
transmit the reduced set of audio features to a cloud computing device in response to the identification of the one or more keywords; and
obtain a response to the reduced set of audio features from the cloud computing device to cause an output.

### Claim 19 (depends on 18)

The wearable apparatus of claim 18, wherein the means for processing includes a key phrase detector and a feature extractor, and wherein the means for communicating includes a transmitter.