Phoneme Alignment Using the Information on Phonological Processes in Continuous Speech

Phoneme Alignment Using the Information on Phonological Processes in Continuous Speech Daniil Kocharov Department of Phonetics, Saint Petersburg State University, 7/9 Universitetskaya nab., St. Petersburg, 199034 Russia kocharov@phonetics.pu.ru Abstract The current study focuses on optimization of Levenshtein algorithm for the purpose of computing the optimal alignment between two phoneme transcriptions of spoken utterance containing sequences of phonetic symbols. The alignment is computed with the help of a confusion matrix in which costs for phonetic symbol deletion, insertion and substitution are defined taking into account various phonological processes that occur in fluent speech, such as anticipatory assimilation, phone elision and epenthesis. The corpus containing about 30 hours of Russian read speech was used to evaluate the presented algorithms. The experimental results have shown significant reduction of misalignment rate in comparison with the baseline Levenshtein algorithm: the number of errors has been reduced from 1.1 % to 0.28 %. Keywords: automatic phoneme alignment, phonological process, confusion matrices, Russian 1. Introduction The goal of the research presented in this paper is to align effectively two sequences of phonemes (phonetic transcrip- tions) that describe the same speech signal. This task is important for sociolinguistic and dialectological research on how various people read or speak the same text, e.g. (Heeringa, 2004), (Valls et al., 2013). Another field of ap- plication is comparison and alignment of annotations pro- duced by different human or machine transcribers for the same speech data, e.g. ( ´Alvarez et al., 2014). The current work has been done as a part of a research on inter-speaker variability. There has been a need in predic- tion of the pronunciation deviation of various Russian na- tive speakers from Standard Russian. A correct compari- son of individual pronunciations requires transcriptions to be perfectly aligned. The linguistic approach assumes that for a correct alignment of phoneme sequences one should consider the behavior of phonemes in continuous speech under different conditions. Automatic aligners using various linguistic approaches have been made for many languages, including English ( ´Alvarez et al., 2014), Spanish (Valls et al., 2013), Dutch (Elffers et al., 2005), Norwegian (Heeringa, 2004), Basque (Bordel et al., 2012). Such an aligner for Russian is pre- sented in the paper. It is based on the usage of phoneme sets defined in such a way that a phoneme is more probable to substitute another phoneme from the same set than from another set. Besides substitutions, the approach suggests taking into account frequent phone insertions and deletions, whereas in previous works only substitution cost was esti- mated. Section 2. presents briefly the information on phonolog- ical processes in Russian speech that have been consid- ered while developing the aligner. Section 3. describes the modifications that are proposed for basic Levenshtein aligner. The evaluation procedure for estimating alignment efficiency is described in section 4.. The achieved results are shown in section 5.. 2. Phonological Processes in Russian Speech There are both context-dependent and context-independent phone alternations, elisions or epentheses in the Russian speech. The majority of these speech events are either …

Content truncated to 500 words