Speech Recognition Device and Speech Recognition Method

• US 20080270127A1
• Filed: 03/15/2005
• Published: 10/30/2008
• Est. Priority Date: 03/31/2004
• Status: Active Grant

• Current Assignee: Pioneer Corporation
• Original Assignee: Pioneer Corporation
• Inventors: Kobayashi, Hajime, Toyama, Soichi, Suzuki, Yasunori
• Granted Patent: US 7,813,921 B2
• Time in Patent Office: Days
• US Class Current: 704/226
• CPC Class Codes: G10L 15/065 Adaptation
• CPC Class Codes: 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
• CPC Class Codes: G10L 2015/0635 updating or merging of old and new templates; Mean values; Weighting

Abstract

There is provided a voice recognition device and a voice recognition method that enhance the function of noise adaptation processing in voice recognition processing and reduce the capacity of a memory being used. Acoustic models are subjected to clustering processing to calculate the centroid of each cluster and the differential vector between the centroid and each model, model composition between each kind of assumed noise model and the calculated centroid is carried out, and the centroid of each composition model and the differential vector are stored in a memory. In the actual recognition processing, the centroid optimal to the environment estimated by the utterance environmental estimation is extracted from the memory, model restoration is carried out on the extracted centroid by using the differential vector stored in the memory, and noise adaptation processing is executed on the basis of the restored model.

Claims

1. A voice recognition device for executing noise adaptation processing based on a noise model on an input voice signal to carry out voice recognition on the input voice signal is characterized by comprising:

first storage means for calculating a first parameter representative of plural noise models contained in each of plural noise environmental categories in advance and storing the first parameter every noise environmental category;

second storage means for calculating a second parameter representing relative positional information between each of the plural noise models and the first parameter in advance and storing the second parameter;

estimating means for estimating, on the basis of the characteristic of an environmental noise superposed by the input voice signal, a noise environmental category to which the environmental noise concerned belongs;

selecting means for selecting and extracting the first parameter corresponding to a noise environmental category estimated by the estimating means from the first storage means; and

noise adaptation means for restoring a noise model adaptive to the environmental noise by using the first parameter extracted by the selecting means and the second parameter read out from the second storage means and executing noise adaptation processing on the input voice signal by using the noise model thus restored.

2. The voice recognition device according to claim 1, wherein the first parameter contains a centroid value associated with each noise model that is achieved by executing model composition processing on a centroid value achieved by executing clustering processing on an acoustic model and each of plural noise models contained in one noise environmental category, and data appended to the centroid value.

3. The voice recognition device according to claim 1, further comprising a non-stationary parameter removing processor for removing non-stationary parameters from a set of feature parameters of an environmental noise superposed on the input voice signal.

4. The voice recognition device according to claim 2, wherein the clustering processing is continued until the number of groups of acoustic models formed by the processing concerned reaches a predetermined group number.

5. The voice recognition device according to claim 2, wherein the second parameter is a differential vector between the centroid value and each of the plural noise models.

6. The voice recognition device according to claim 1, wherein the estimating means further comprises storing and adding means for extracting the first parameter from the environmental noise and adding and storing the first parameter to the first storage means when it is detected that the environmental noise does not corresponds to a noise environmental category prepared in advance.

7. The voice recognition device according to claim 1, further comprising communication means for relaying data between a server containing a data base and a memory contained in the first and second storage means, wherein the data base is used as a part or the whole of the memory.

8. A voice recognition method for executing noise adaptation processing based on a noise model on an input voice signal to carry out voice recognition on the input voice signal is characterized by comprising:

a step of calculating a first parameter representative of plural noise models contained in each of plural noise environmental categories in advance and storing the first parameter into a first memory every noise environmental category;

a step of calculating a second parameter representing relative positional information between each of the plural noise models and the first parameter in advance and storing the second parameter into a second memory;

a step of estimating, on the basis of the characteristic of an environmental noise superposed by the input voice signal, a noise environmental category to which the environmental noise concerned belongs;

a step of selecting and extracting the first parameter corresponding to an estimated noise environmental category from the first memory; and

a step of restoring a noise model adaptive to the environmental noise by using the selected and extracted first parameter and the second parameter read out from the second memory and executing noise adaptation processing on the input voice signal by using the noise model thus restored.

9. The voice recognition method according to claim 8, further comprising a step of removing non-stationary parameters from a set of feature parameters of an environmental noise superposed on the input voice signal.