# US 7,613,611 B2 · Method and apparatus for vocal-cord signal recognition

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

- Number: US 7,613,611 B2
- Document ID: US7613611B2
- Status: Granted
- Kind: B2
- Application number: 11/140,105
- Filed: 2005-05-26
- Published: 2006-05-04
- Granted: 2009-11-03
- Earliest claimed: 2004-11-04
- Assignees: Electronics and Telecommunications Research Institute
- Inventors: Kwan Hyun Cho; Mun Sung Han; Young Giu Jung; Hee Sook Shin; Jun Seok Park; Dong Won Han
- Web page: https://patentgrep.com/patent/US7613611B2

## CPC

- G10L 15/10: using distance or distortion measures between unknown speech and reference templates
- G10L 15/24: Speech recognition using non-acoustical features
- G10L 25/93: Discriminating between voiced and unvoiced parts of speech signals (G10L25/90 takes precedence)
- G10L 2021/02168: {the estimation exclusively taking place during speech pauses}
- G10L: SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING

## Abstract

Provided is a method and an apparatus for vocal-cord signal recognition. A signal processing unit receives and digitalizes a vocal cord signal, and a noise removing unit which channel noise included in the vocal cord signal. A feature extracting unit extracts a feature vector from the vocal cord signal, which has the channel noise removed therefrom, and a recognizing unit calculates a similarity between the vocal cord signal and the learned model parameter. Consequently, the apparatus is robust in a noisy environment.

## Claims

### Claim 1

An apparatus for vocal-cord signal recognition, comprising:
a signal processing unit which receives a vocal cord signal and digitalizes the vocal cord signal;
a noise removing unit which removes channel noise included in the vocal cord signal, the noise removing unit to calculate an average cepstrum of a predetermined mute section of the vocal cord signal, and to subtract the average cepstrum from a cepstrum of each frame of the vocal cord signal;
a feature extracting unit which extracts a feature vector from the vocal cord signal, which has the channel noise removed therefrom; and
a recognizing unit which calculates a similarity between the vocal cord signal and a learned model parameter.

### Claim 2 (depends on 1)

The apparatus of claim 1, wherein the signal processing unit calculates a threshold for re-adjusting a gain appropriate for an end point detection of the vocal cord signal in a wireless receiver, which receives the vocal cord signal through a wireless personal area network, and transmits the calculated threshold to a wireless transmitter to re-adjust a gain of an amplifier.

### Claim 3 (depends on 1)

The apparatus of claim 1, wherein the noise removing unit uses {circumflex over (X)}{circumflex over (Xt)}=Xt−Nnew, which subtracts local moving average noise cepstrum renewed by Nnew=α×Nold+(1−α)×Nc from cepstrum of each frame,
wherein α is a weight, Nc is a noise cepstrum of a first time period, Nold is a noise cepstrum of a second time period preceding the first time period, and Xt and {circumflex over (X)}{circumflex over (Xt)} is a cepstrum of an original signal of a current frame and a cepstrum with noise removed, respectively.

### Claim 4 (depends on 1)

The apparatus of claim 1, wherein the feature extracting unit detects a signal section of the vocal cord signal using a method of end point detection through dispersion of magnitude in a frequency domain, and extracts a feature vector from the vocal cord signal of a detected signal section.

### Claim 5 (depends on 3)

The apparatus of claim 3, wherein the feature extracting unit extracts a real-time feature vector from a voice section, a start point and an end point of which are determined by (St≧BaseThreshold) and (abs(St−Nt)≧RelativeThreshold) and (St≦BaseThreshold) and (ContinuousNoiseFrames≧NoiseDuration), respectively,
wherein,
S
t
=
1
N
1
⁢
∑
i
=
t
-
N
1
t
⁢
X
i
,
N
t
=
1
N
2
⁢
∑
i
=
t
-
N
1
-
N
2
t
-
N
1
⁢
X
i
Xi is a spectrum distribution when t=i, BaseThreshold is a minimum limit value of the vocal cord signal, RelativeThreshold is a threshold for comparing a relative difference between St and Nt, and NoiseDuration is a value for determining the end point of the vocal cord signal.

### Claim 6 (depends on 1)

The apparatus of claim 1, wherein the signal processing unit comprises:
a vocal cord microphone which receives and digitalizes the vocal cord signal;
a wireless transmitter which transmits the vocal cord signal through wireless communication; and
a wireless receiver which receives the vocal cord signal output from the wireless transmitter.

### Claim 7

A method of speech recognition, comprising:
receives a vocal cord signal through a neck microphone;
removing channel noise included in the vocal cord signal by calculating an average cepstrum of a predetermined mute section of the vocal cord signal and subtracting the average cepstrum from a cepstrum of each frame of the vocal cord signal;
extracting a feature vector from the vocal cord signal, which has the channel noise removed therefrom; and
recognizing speech by calculating similarity between the vocal cord signal and a learned model parameter.

### Claim 8 (depends on 7)

The method of claim 7, wherein the extracting of the feature vector from the vocal cord signal comprises:
detecting a signal section of the vocal cord signal using a method of end point detection dispersion of magnitude in a frequency domain; and
extracting a feature-vector from the vocal cord signal of the detected signal section.