# US 2018/0025730 A1 · Circuit and method for speech recognition

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

- Number: US 2018/0025730 A1
- Document ID: US20180025730A1
- Status: Published application
- Kind: A1
- Application number: 15/654,153
- Filed: 2017-07-19
- Published: 2018-01-25
- Earliest claimed: 2016-07-22
- Assignees: Dolphin Integration
- Applicants: Dolphin Integration
- Inventors: Paul GILETTI
- Web page: https://patentgrep.com/patent/US20180025730A1

## CPC

- G10L 17/00: Speaker identification or verification techniques
- G10L 15/142: {Hidden Markov Models [HMMs]}
- G10L 15/16: using artificial neural networks
- G10L 15/20: Speech recognition techniques specially adapted for robustness in adverse environments, e.g. in noise, of stress induced speech (G10L21/02 takes precedence)
- 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 25/84: for discriminating voice from noise
- G10L 2015/025: {Phonemes, fenemes or fenones being the recognition units}
- G10L 2025/783: {based on threshold decision}
- G10L 2025/786: {Adaptive threshold}
- G10L: SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING

## Abstract

The invention concerns a circuit for speech recognition comprising: a voice detection circuit configured to detect, based on at least one input parameter, the presence of a voice signal in an input audio signal and to generate an activation signal on each voice detection event; a speech recognition circuit configured to be activated by the activation signal and to perform speech recognition on the input audio signal, the speech recognition circuit being further configured to generate an output signal indicating, based on the speech recognition, whether each voice detection event is true or false; and an analysis circuit configured to generate, based on the output signal of the speech recognition circuit, a control signal for modifying one or more of said input parameters.

## Claims

### Claim 1

A circuit for speech recognition comprising:
a voice detection circuit configured to detect, based on at least one input parameter, the presence of a voice signal in an input audio signal and to generate an activation signal on each voice detection event;
a speech recognition circuit configured to be activated by the activation signal and to perform speech recognition on the input audio signal, the speech recognition circuit being further configured to generate an output signal indicating, based on the speech recognition, whether each voice detection event is true or false; and
an analysis circuit configured to:
generate a first value representing a true detection rate indicating the rate of true detection events among the total number of detection events generated by the voice detection circuit and/or a second value representing a false detection rate indicating the rate of false detection events among the total number of detection events generated by the voice detection circuit; and
generate a control signal for modifying one or more of said input parameters based on a comparison of the first value with a first threshold or of the second value with a second threshold and/or based on whether the first or second value is increasing or decreasing.

### Claim 2 (depends on 1)

The circuit of claim 1, wherein the analysis circuit is further configured to:
generate a third value representing the total number of detection events generated by the voice detection circuit; and
compare the third value with a third threshold and/or determine whether the third value is increasing or decreasing.

### Claim 3 (depends on 2)

The circuit of claim 2, wherein the analysis circuit is configured to modify the control signal to render the voice detection circuit more sensitive if the first threshold and/or the third threshold is not exceeded and/or if the first value and/or third value is decreasing.

### Claim 4 (depends on 3)

The circuit of claim 3, wherein the analysis circuit is further configured to modify the control signal to render the voice detection circuit less sensitive if the second and/or third threshold is exceeded and/or if the second value is increasing.

### Claim 5 (depends on 1)

The circuit of claim 1, wherein each of the at least one input parameter is one of the following parameters:
a parameter setting one or more cut-off frequencies of one or more filters of the voice detection circuit;
a parameter setting one or more detection thresholds of the voice detection circuit;
a parameter setting one or more energy computation periods;
a parameter setting a threshold number of zero passages of the audio signal;
a parameter setting one or more debounce times; and
a parameter setting a weighting to be applied to the output signal of one or more sub-circuits of the voice detection circuit.

### Claim 6 (depends on 1)

The circuit of claim 1, wherein the speech recognition circuit is adapted to generate a further activation signal for activating one or more further circuits in response to a speech recognition event.

### Claim 7 (depends on 6)

The circuit of claim 6, wherein the speech recognition circuit is adapted to perform keyword recognition.

### Claim 8

A method of speech recognition comprising:
detecting, by a voice detection circuit based on at least one input parameter, the presence of a voice signal in an input audio signal;
generating, by the voice detection circuit an activation signal on each voice detection event;
activating a speech recognition circuit by the activation signal;
performing, by the speech recognition circuit, speech recognition on the input audio signal;
generating, by the speech recognition circuit, an output signal indicating, based on the speech recognition, whether each voice detection event is true or false; generating, by the analysis circuit, a first value representing a true detection rate indicating the rate of true detection events among the total number of detection events generated by the voice detection circuit and/or a second value representing a false detection rate indicating the rate of false detection events among the total number of detection events generated by the voice detection circuit; and
generating, by an analysis circuit, a control signal for modifying one or more of said input parameters based on a comparison of the first value with a first threshold value or of the second value with a second threshold value and/or on whether the first or second value is increasing or decreasing.
