Abstract: This paper presents an approach to automatic segmentation of speech corpora. The availability of sufficiently precise labelled sentences can avoid the need for a segmentation by human experts. The goal of this process is to prepare speech corpora both for training acoustic models and for concatenative text to speech synthesis. Our system only needs the speech signal and the phonetic sequence for each sentence of a corpus. It estimates a GMM by using all sentences, where each Gaussian distribution represents an acoustic class. Then it combines the probability densities of each acoustic class with a set of conditional probabilities in order to estimate the probability densities of the states of each phonetic unit. A DTW algorithm fixes the phonetic boundaries using the known phonetic sequence. This DTW is a step inside an iterative process which aims to segment the corpus and re-estimate the conditional probabilities. A flat start setup is used to give initial values to the conditional probabilities.
Index Terms: automatic speech segmentation, phoneme boundaries detection, phoneme alignment.