A Minimum Boundary Error Framework for Automatic Phonetic Segmentation

Abstract

Abstract. This paper presents a novel framework for HMM-based automatic phonetic segmentation that improves the accuracy of placing phone boundaries. In the framework, both training and segmentation approaches are proposed according to the minimum boundary error (MBE) criterion, which tries to minimize the expected boundary errors over a set of possible phonetic alignments. This framework is inspired by the recently proposed minimum phone error (MPE) training approach and the minimum Bayes risk decoding algorithm for automatic speech recognition. To evaluate the proposed MBE framework, we conduct automatic phonetic segmentation experiments on the TIMIT acoustic-phonetic continuous speech corpus. MBE segmentation with MBE-trained models can identify 80.53% of human-labeled phone boundaries within a tolerance of 10 ms, compared to 71.10% identified by conventional ML segmentation with ML-trained models. Moreover, by using the MBE framework, only 7.15% of automatically labeled phone boundaries have errors larger than 20 ms.

Comparison of forced-alignment speech recognition and humans for generating reference VAD

“…ASR systems are extensively used for the initial segmentation of speech. A HMM based phonetic recognizer is commonly employed for phoneme segmentation and for estimating the phoneme boundaries by means of Viterbi forced-alignment [5], [6].…”

Framework for cross-language automatic phonetic segmentation

“…The state of the art automatic segmentation systems are mainly HMM-based [1,2]. Boundaries produced by HMMs are subject to bias errors due to a range of factors, mainly, the training algorithms and the minimum duration of HMMs.…”

Improvements on Automatic Speech Segmentation at the Phonetic Level

“…An approach inspired in the minimum phone error training algorithm for automatic speech recognition [9] is presented in [10]. The objective of this approach is to minimize the expected boundary errors over a set of phonetic alignments represented as a phonetic lattice.…”