Improved tone modeling for Mandarin broadcast news speech recognition

Published 17 September 2006

Xin Lei, Man-Hung Siu, Mei-Yuh Hwang, Mari Ostendorf, Tan Lee

DOI10.21437/interspeech.2006-372

Citations88

TL;DR

This paper combines the embedded aproach (using improved F0 smoothing) with explicit tone modeling in rescoring the output lattices of Mandarin automatic speech recognition systems to improve the character error rate on the CTV test set.

Abstract

Tone has a crucial role in Mandarin speech in distinguishing ambiguous words. Most state-of-the-art Mandarin automatic speech recognition systems adopt embedded tone modeling, where tonal acoustic units are used and F0 features are appended to the spectral feature vector. In this paper, we combine the embedded aproach (using improved F0 smoothing) with explicit tone modeling in rescoring the output lattices. Oracle experiments indicate 32% relative improvement can be achieved by rescoring with perfect tone information. Recognition experiments on Mandarin broadcast news show that, even with an accuracy of only 70%, the explicit tone classifier offers complementary knowledge and improves performance significantly. Through the combination of tone modeling techniques, the character error rate on the CTV test set can be improved from 13.0% to 11.5%.

Keywords

Computer Science

Large vocabulary Mandarin speech recognition with different approaches in modeling tones

79 Citations2000Eric Chang, Jian-Lai Zhou +3 more

A large vocabulary Mandarin speech recognition system based on Microsoft’s Whisper system is described and it is suggested that other sources of information such as energy and duration can also contribute toward disambiguating between different tones.

IEEE Transactions on Speech and Audio ProcessingTone recognition of continuous Mandarin speech based on neural networks

67 Citations1995Sim-Horng Chen, Yih‐Ru Wang

Two approaches using HCNN and HSMLP to model the intonation pattern as a hidden Markov chain for assisting tone recognition are proposed, and the effectiveness of these schemes was confirmed by simulations on a speaker-independent tone recognition task.

Pitch tracking and tone features for Mandarin speech recognition

60 Citations2002Hank Huang, Frank Seide

An efficient real-time pitch tracker and a set of tone features that achieve a vast 30% reduction of the character error rate (CER) compared to the non-tonal baseline, to be the highest improvement from tones ever reported for Mandarin.

ACM Transactions on Asian Language Information ProcessingUsing tone information in Cantonese continuous speech recognition

53 Citations2002Tan Lee, Wai Pik Lau +2 more

Experimental results show that reliable tone information helps to improve the overall performance of LVCSR, the automatic speech recognition of Chinese.

Speech CommunicationTone recognition of continuous Cantonese speech based on support vector machines

53 Citations2004Gang Peng, William Yang Wang

This paper looks at using support vector machines for automatic tone recognition in continuously spoken Cantonese, which is well known for its complex tone system, and proposes an adaptive log-scale 5-level F 0 normalization method to reduce the tone-irrelevant variation of F 0 values.

DSpace@MIT (Massachusetts Institute of Technology)Prosodic modeling for improved speech recognition and understanding

24 Citations2001Chao Wang, Stephanie Seneff

It is demonstrated that tone classification performance can be significantly improved by taking into account F0 declination, phrase boundary, and tone context influences, and various ways to incorporate tone model constraints into the SUMMIT speech recognition system are explored.

View PDF

Decision tree based tone modeling for Chinese speech recognition

17 Citations2004Pui-Fung WONG, Man-Hung Siu

A robust polynomial segmental representation of the pitch contour coupled with a decision tree based tone classifier is proposed that gave a tone classification-accuracy of 89% for isolated syllables and 71.2% for the continuous speech.

Use of tone information in cantonese lvcsr based on generalized character posterior probability decoding

9 Citations2005Tan Lee, Yao Qian

A decoding algorithm is developed that can minimize the character error rate and is applied to a reduced search space, e.g. a word graph or the N-best sentence list, which results from the 1st pass of search, and the generalized character posterior probability (GCPP) is maximized.

Porting Decipher from English to Mandarin

8 Citations2006Mei-Yuh Hwang, Xin Lei +6 more

This paper describes the efforts in porting the SRI Decipher English system into Mandarin for transcribing telephone conversations, including all aspects of the system: the pronunciation phone set and lexicon, word segmentation, pitch features, discriminatively trained acoustic models with parameter sharing determined by decision trees, and web-data augmented language models.