Continuous Speech Recognition Based on Deterministic Finite Automata Machine using Utterance and Pitch Verification

TL;DR

This paper presents novel acoustic modeling techniques for utterance verification in continuous speech recognition, utilizing likelihood ratios and finite-state transducers to significantly improve verification accuracy.

Contribution

It introduces two UV techniques based on likelihood ratios and pitch recognition with finite-state transducers, enhancing UV performance in continuous speech recognition systems.

Findings

Likelihood ratio based parameter estimation improves UV performance

Use of weighted finite-state transducers yields up to 11% accuracy improvement

Finite state machine performance depends heavily on LR criterion in training

Abstract

This paper introduces a set of acoustic modeling techniques for utterance verification (UV) based continuous speech recognition (CSR). Utterance verification in this work implies the ability to determine when portions of a hypothesized word string correspond to incorrectly decoded vocabulary words or out-of-vocabulary words that may appear in an utterance. This capability is implemented here as a likelihood ratio (LR). There are two UV techniques that are presented here. The first is voice verification along with the vocabulary testing, at the same time the parameters for UV models are generated based on the optimization criterion which is directly related to the LR measure. The second technique is a pitch recognition based on weighted finite-state transducers. These transducers provide a common and natural representation for major components of speech recognition systems, including hidden Markov models (HMMs), context-dependency models, pronunciation dictionaries, statistical grammars, and word or phone lattices. The finite state machine processes the acoustic parameters of UV models. The results of an experimental study presented in the paper shows that LR based parameter estimation results in a significant improvement in UV performance for this task. The study also found that the use of the LR based weighted finite-state transducers along with the UV, can provide as much as an 11% improvement in UV performance when compared to existing UV procedures. Finally, it was also found that the performance of the finite state machine was highly dependent on the use of the LR criterion in training acoustic models. Several observations are made in the paper concerning the formation of confidence measures for UV and the interaction of these techniques with statistical language models.