Speech-processing apparatus and speech-processing method

• US 10,002,623 B2
• Filed: 07/29/2016
• Issued: 06/19/2018
• Est. Priority Date: 09/28/2015
• Status: Active Grant

• Current Assignee: Honda Motor Co., Ltd. (Honda Motor Company)
• Original Assignee: Honda Motor Co., Ltd. (Honda Motor Company)
• Inventors: Nakamura, Keisuke, Nakadai, Kazuhiro
• Primary Examiner(s): Pham, Thierry L
• Application Number:
  US15/223,478
• Publication Number:
  US 20170092298A1
• Time in Patent Office: 690 Days
• Field of Search: 704226, 704233, 704235, 704246
• CPC Class Codes: G01S 3/80 using ultrasonic, sonic or infrasonic waves
• CPC Class Codes: G10K 11/1754 Speech masking
• CPC Class Codes: G10L 2025/783 based on threshold decision
• CPC Class Codes: G10L 21/0272 Voice signal separating
• CPC Class Codes: G10L 25/18 the extracted parameters being spectral information of each sub-band
• CPC Class Codes: G10L 25/84 for discriminating voice from noise
• CPC Class Codes: G10L 25/93 Discriminating between voiced and unvoiced parts of speech signals G10L25/90 takes precedence

Abstract

A speech-processing apparatus includes: a sound source localization unit that localizes a sound source based on an acquired speech signal; and a speech zone detection unit that performs speech zone detection based on localization information localized by the sound source localization unit.

Claims

1. A speech-processing apparatus, comprising:

a processor configured to;

localize a sound source based on an acquired speech signal; and

perform speech zone detection in which a speech start and a speech end are detected based on localization information of the localized sound source,

wherein the processor is configured to perform the speech zone detection by using a plurality of threshold values with respect to the localized speech signal, and wherein

the processor is configured to;

detect a sound source candidate by using a first threshold value of the plurality of threshold values with respect to the localized speech signal,

perform a clustering process on the detected sound source candidate, and

perform the speech zone detection in which a speech start and a speech end are detected, by using a second threshold value that is larger than the first threshold value of the plurality of threshold values for each cluster classified by the clustering process.

2. The speech-processing apparatus according to claim 1, wherein the processor is configured to:

detect a sound source candidate by using a second threshold value of the plurality of threshold values with respect to the localized speech signal,

perform a clustering process on the detected sound source candidate, and

perform the speech zone detection in which a speech start and a speech end are detected, by using a first threshold value that is smaller than the second threshold value of the plurality of threshold values for each cluster classified by the clustering process.

3. The speech-processing apparatus according to claim 1, wherein the processor is configured to perform the speech zone detection in which a speech start and a speech end are detected, based on a gradient of a spatial spectrum of the localized speech signal.

4. The speech-processing apparatus according to claim 1, wherein the processor is configured to:

perform sound source separation based on the acquired speech signal;

perform sound source identification based on the separated separation signal; and

detect, when the identified result is speech, that speech is continued in a zone.

5. The speech-processing apparatus according to claim 1, wherein the processor is configured to:

detect a sound source candidate by using a threshold value with respect to the localized speech signal,

acquire event information indicating that an event which causes noise with respect to the speech signal is occurring,

generate a mask for a sound source candidate detected by using the threshold value based on the acquired event information, and

perform the speech zone detection in which a speech start and a speech end are detected, by using the mask generated for the sound source candidate.

6. A speech-processing method, comprising:

(a) localizing a sound source based on an acquired speech signal;

(b) performing speech zone detection in which a speech start and a speech end are detected based on localization information of the sound source localized in (a); and

(c) performing the speech zone detection by using a plurality of threshold values with respect to the speech signal localized in (a), wherein in (c),

a sound source candidate is detected by using a first threshold value of the plurality of threshold values with respect to the localized speech signal,

a clustering process is performed on the detected sound source candidate, and

the speech zone detection in which a speech start and a speech end are detected is performed by using a second threshold value that is larger than the first threshold value of the plurality of threshold values for each cluster classified by the clustering process.