Utterance-Based Selective Training for the Automatic Creation of Task-Dependent Acoustic Models

Tobias CINCAREK; Tomoki TODA; Hiroshi SARUWATARI; Kiyohiro SHIKANO

doi:10.1093/ietisy/e89-d.3.962

Utterance-Based Selective Training for the Automatic Creation of Task-Dependent Acoustic Models

Tobias CINCAREK, Tomoki TODA, Hiroshi SARUWATARI, Kiyohiro SHIKANO

Full Text Views

0

Cite this

Summary :

To obtain a robust acoustic model for a certain speech recognition task, a large amount of speech data is necessary. However, the preparation of speech data including recording and transcription is very costly and time-consuming. Although there are attempts to build generic acoustic models which are portable among different applications, speech recognition performance is typically task-dependent. This paper introduces a method for automatically building task-dependent acoustic models based on selective training. Instead of setting up a new database, only a small amount of task-specific development data needs to be collected. Based on the likelihood of the target model parameters given this development data, utterances which are acoustically close to the development data are selected from existing speech data resources. Since there are too many possibilities for selecting a data subset from a larger database in general, a heuristic has to be employed. The proposed algorithm deletes single utterances temporarily or alternates between successive deletion and addition of multiple utterances. In order to make selective training computationally practical, model retraining and likelihood calculation need to be fast. It is shown, that the model likelihood can be calculated fast and easily based on sufficient statistics without the need for explicit reconstruction of model parameters. The algorithm is applied to obtain an infant- and elderly-dependent acoustic model with only very few development data available. There is an improvement in word accuracy of up to 9% in comparison to conventional EM training without selection. Furthermore, the approach was also better than MLLR and MAP adaptation with the development data.

Publication: IEICE TRANSACTIONS on Information Vol.E89-D No.3 pp.962-969

Publication Date: 2006/03/01

Publicized

Online ISSN: 1745-1361

DOI: 10.1093/ietisy/e89-d.3.962

Type of Manuscript: Special Section PAPER (Special Section on Statistical Modeling for Speech Processing)

Category: Speech Recognition

Cite this

Copy

Tobias CINCAREK, Tomoki TODA, Hiroshi SARUWATARI, Kiyohiro SHIKANO, "Utterance-Based Selective Training for the Automatic Creation of Task-Dependent Acoustic Models" in IEICE TRANSACTIONS on Information, vol. E89-D, no. 3, pp. 962-969, March 2006, doi: 10.1093/ietisy/e89-d.3.962.
Abstract: To obtain a robust acoustic model for a certain speech recognition task, a large amount of speech data is necessary. However, the preparation of speech data including recording and transcription is very costly and time-consuming. Although there are attempts to build generic acoustic models which are portable among different applications, speech recognition performance is typically task-dependent. This paper introduces a method for automatically building task-dependent acoustic models based on selective training. Instead of setting up a new database, only a small amount of task-specific development data needs to be collected. Based on the likelihood of the target model parameters given this development data, utterances which are acoustically close to the development data are selected from existing speech data resources. Since there are too many possibilities for selecting a data subset from a larger database in general, a heuristic has to be employed. The proposed algorithm deletes single utterances temporarily or alternates between successive deletion and addition of multiple utterances. In order to make selective training computationally practical, model retraining and likelihood calculation need to be fast. It is shown, that the model likelihood can be calculated fast and easily based on sufficient statistics without the need for explicit reconstruction of model parameters. The algorithm is applied to obtain an infant- and elderly-dependent acoustic model with only very few development data available. There is an improvement in word accuracy of up to 9% in comparison to conventional EM training without selection. Furthermore, the approach was also better than MLLR and MAP adaptation with the development data.
URL: https://global.ieice.org/en_transactions/information/10.1093/ietisy/e89-d.3.962/_p

Copy

@ARTICLE{e89-d_3_962,
author={Tobias CINCAREK, Tomoki TODA, Hiroshi SARUWATARI, Kiyohiro SHIKANO, },
journal={IEICE TRANSACTIONS on Information},
title={Utterance-Based Selective Training for the Automatic Creation of Task-Dependent Acoustic Models},
year={2006},
volume={E89-D},
number={3},
pages={962-969},
abstract={To obtain a robust acoustic model for a certain speech recognition task, a large amount of speech data is necessary. However, the preparation of speech data including recording and transcription is very costly and time-consuming. Although there are attempts to build generic acoustic models which are portable among different applications, speech recognition performance is typically task-dependent. This paper introduces a method for automatically building task-dependent acoustic models based on selective training. Instead of setting up a new database, only a small amount of task-specific development data needs to be collected. Based on the likelihood of the target model parameters given this development data, utterances which are acoustically close to the development data are selected from existing speech data resources. Since there are too many possibilities for selecting a data subset from a larger database in general, a heuristic has to be employed. The proposed algorithm deletes single utterances temporarily or alternates between successive deletion and addition of multiple utterances. In order to make selective training computationally practical, model retraining and likelihood calculation need to be fast. It is shown, that the model likelihood can be calculated fast and easily based on sufficient statistics without the need for explicit reconstruction of model parameters. The algorithm is applied to obtain an infant- and elderly-dependent acoustic model with only very few development data available. There is an improvement in word accuracy of up to 9% in comparison to conventional EM training without selection. Furthermore, the approach was also better than MLLR and MAP adaptation with the development data.},
keywords={},
doi={10.1093/ietisy/e89-d.3.962},
ISSN={1745-1361},
month={March},}

Copy

TY - JOUR
TI - Utterance-Based Selective Training for the Automatic Creation of Task-Dependent Acoustic Models
T2 - IEICE TRANSACTIONS on Information
SP - 962
EP - 969
AU - Tobias CINCAREK
AU - Tomoki TODA
AU - Hiroshi SARUWATARI
AU - Kiyohiro SHIKANO
PY - 2006
DO - 10.1093/ietisy/e89-d.3.962
JO - IEICE TRANSACTIONS on Information
SN - 1745-1361
VL - E89-D
IS - 3
JA - IEICE TRANSACTIONS on Information
Y1 - March 2006
AB - To obtain a robust acoustic model for a certain speech recognition task, a large amount of speech data is necessary. However, the preparation of speech data including recording and transcription is very costly and time-consuming. Although there are attempts to build generic acoustic models which are portable among different applications, speech recognition performance is typically task-dependent. This paper introduces a method for automatically building task-dependent acoustic models based on selective training. Instead of setting up a new database, only a small amount of task-specific development data needs to be collected. Based on the likelihood of the target model parameters given this development data, utterances which are acoustically close to the development data are selected from existing speech data resources. Since there are too many possibilities for selecting a data subset from a larger database in general, a heuristic has to be employed. The proposed algorithm deletes single utterances temporarily or alternates between successive deletion and addition of multiple utterances. In order to make selective training computationally practical, model retraining and likelihood calculation need to be fast. It is shown, that the model likelihood can be calculated fast and easily based on sufficient statistics without the need for explicit reconstruction of model parameters. The algorithm is applied to obtain an infant- and elderly-dependent acoustic model with only very few development data available. There is an improvement in word accuracy of up to 9% in comparison to conventional EM training without selection. Furthermore, the approach was also better than MLLR and MAP adaptation with the development data.
ER -