Marking the Allophones Boundaries Based on the DTW Algorithm

摘要

The paper presents an approach to marking the boundaries of allophones in the speech signal based on the Dynamic Time Warping (DTW) algorithm. Setting and marking of allophones boundaries in continuous speech is a difficult issue due to the mutual influence of adjacent phonemes on each other. It is this neighbourhood on the one hand that creates variants of phonemes that is allophones, and on the other hand it affects that the border between allophones is in some cases very difficult to determine. Nowadays, this task is carried out manually in cooperation with specialists in the field of phonetics. The presented approach allows to build a system that is able to automate this process. The aim of the work currently carried out by the author is a method that facilitates the training material processing for the needs of the development of multimodal speech recognition systems. For this purpose, the difficult problem of marking boundaries of allophones is solved in this report based on the Polish dictionary in the context of the creation of allophone bases for speech synthesis. This is done in this way due to the simplified possibility of organizing critical listening and subjective evaluation of received allophones by a large group of Polish native speakers (73 people). Strengthening the method will allow it to be used for the extraction of allophones for the needs of developed system of automatic transcription of English speech and for its notation according to the IPA standard. The analysed continuous speech is combined in the DTW algorithm with a synthesized speech signal. The comparison of both signals is performed not in the time domain as in the classical DTW, but in the frequency domain. This allows for a statement that the phonetic content of both signals is compared. The paper describes the process of marking the boundaries of allophones for the Polish language, however after appropriate modifications, this approach can be used to determine the allophones boundaries in other languages, especially for English.