Rapid speaker adaptation using speaker-mixture allophone models applied to speaker-independent speech recognition

Kosaka, Tetsuo; Takami, Jun-ichi; Sagayama, Shigeki

doi:10.1109/icassp.1993.319371

Cited by 8 publications

(4 citation statements)

References 8 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We have already proposed a speaker-mixture method (Kosaka et al, 1993) that can yield highly accurate speaker-independent phone models. The same method is also used for recognition in this paper.…”

Section: Recognition Using a Speaker-mixture Modelmentioning

confidence: 99%

See 1 more Smart Citation

Speaker-independent speech recognition based on tree-structured speaker clustering

Kosaka

Matsunaga

Sagayama

1996

Computer Speech & Language

View full text Add to dashboard Cite

Section: Recognition Using a Speaker-mixture Modelmentioning

confidence: 99%

“…HMnets belonging to the selected node are merged to form a K-mixture HMnet by using a speaker-mixture method (Kosaka, Takami & Sagayama, 1993). This HMnet is used for recognition.…”

Section: Recognition Modementioning

confidence: 99%

Speaker-independent speech recognition based on tree-structured speaker clustering

Kosaka

Matsunaga

Sagayama

1996

Computer Speech & Language

View full text Add to dashboard Cite

“…In contrast to the above schemes, the adaptation scheme described here is based on the fact that the training data contains a number of training speakers, some of whom are closer, acoustically, to the test speaker, than the others [10]. 1 If the model parameters are reestimated from the subset of training speakers who are acoustically close to the test speaker, they should be reasonably close to the speakerdependent parameters that would be obtained by training on large amounts of data from the test speaker (if such data were available) [13], [14]. 2 1 Some similar ideas have recently been reported in [11] and [12].…”

Section: B Review Of Adaptation Techniquesmentioning

confidence: 99%

Speaker clustering and transformation for speaker adaptation in speech recognition systems

Padmanabhan

Bahl

Nahamoo

et al. 1998

IEEE Trans. Speech Audio Process.

View full text Add to dashboard Cite

A speaker adaptation strategy is described that is based on finding a subset of speakers, from the training set, who are acoustically close to the test speaker, and using only the data from these speakers (rather than the complete training corpus) to reestimate the system parameters. Further, a linear transformation is computed for every one of the selected training speakers to better map the training speaker's data to the test speaker's acoustic space. Finally, the system parameters (Gaussian means) are reestimated specifically for the test speaker using the transformed data from the selected training speakers. Experiments showed that this scheme is capable of providing an 18% relative improvement in the error rate on a large-vocabulary task with the use of as little as three sentences of adaptation data.

show abstract

“…The SSS-LR achieved good performance in the 1, OOOword recognition experiments, probably due to the high accuracy of the m e t -b a s e d representation. Speaker-independent experiments based on the speaker-mixture HMnet [15] have also been tried. Each mixture component is derived from a particular speaker, and training speech data are used to determine the speaker-mixing weights.…”

Section: Discussionmentioning

confidence: 99%

Speech recognition for Japanese spoken language

Furui

Proceedings of ICSIPNN '94. International Conference on Speech, Image Processing and Neural Networks

View full text Add to dashboard Cite

This paper overviews the speech recognition issues for the Japanese language, and introduces three a n t research projects conducted at NTT (Nippon Telegraph and Telephone) Human Interface Laboratories and ATR Interpreting Telephony Research Laboratories. The fist topic is stochastic language models for sequences of Japanese characters to be used in a Japanese dictation system with unlimited vocabulary. The second topic is an accurate and efficient algorithm for verylarge-vocabulary continuous speech recognition based on an HMM-LR algorithm. This algorithm was applied to a telephone directory assistance system that recognizes spontaneous speech having the vocabulary size of roughly 80, OOO. The third topic is a continuous speech recognition system based on strategies of the phoneme-context-dependent (allophonic) modeling and parsing.

show abstract

Rapid speaker adaptation using speaker-mixture allophone models applied to speaker-independent speech recognition

Cited by 8 publications

References 8 publications

Speaker-independent speech recognition based on tree-structured speaker clustering

Speaker-independent speech recognition based on tree-structured speaker clustering

Speaker clustering and transformation for speaker adaptation in speech recognition systems

Speech recognition for Japanese spoken language

Contact Info

Product

Resources

About