Tree-based state tying for high accuracy acoustic modelling

S.J. Young, J. J. Odell, Philip C. Woodland

1994 · 671 citations · 8 references

DOIFull text

Open access

Concepts

TL;DR

Balancing model complexity with limited training data is a key challenge for HMM-based continuous speech recognizers, especially in large-vocabulary, context-dependent systems where many triphone contexts are unseen. The paper proposes a method to build a tied-state continuous speech recognition system via a phonetic decision tree. The method constructs a phonetic decision tree to tie states across triphones, enabling mapping for unseen contexts. Tree-based clustering achieves recognition performance comparable to earlier data-driven methods while providing mappings for unseen triphones, and outperforms traditional model-based tying, as demonstrated on Resource Management and Wall Street Journal tasks.

Abstract

The key problem to be faced when building a HMM-based continuous speech recogniser is maintaining the balance between model complexity and available training data. For large vocabulary systems requiring cross-word context dependent modelling, this is particularly acute since many such contexts will never occur in the training data. This paper describes a method of creating a tied-state continuous speech recognition system using a phonetic decision tree. This tree-based clustering is shown to lead to similar recognition performance to that obtained using an earlier data-driven approach but to have the additional advantage of providing a mapping for unseen triphones. State-tying is also compared with traditional model-based tying and shown to be clearly superior. Experimental results are presented for both the Resource Management and Wall Street Journal tasks.

References

8