Statistical Analysis of Vietnamese Dialect Corpus and Dialect Identification Experiments

The performance of speech recognition systems will be improved if the corpus is organized in the specialized domain and is applied in a consistent way for speech recognition in specific situations. Vietnamese dialects are various. The building of corpus for Vietnamese dialect is the first step for implementing the system of dialect identification used for increasing the performance of Vietnamese recognition in general. This paper presents a method of building a corpus for Vietnamese dialect identification. Vietnamese corpus VDSPEC is built with topic-based recording and tonal balance. The duration of the corpus is 45.12 hours in total. The basic characteristics and preliminary evaluations of the corpus are also described. The statistical analysis of F0 variation and experiments on the classification of dialects using LDA projection showed that there are distinctions of pronunciation modality of Vietnamese for three dialects Hanoi, Hue and Ho Chi Minh city. For experiments on Vietnamese dialect identification, the first four formants, their bandwidths, and F0 variants have been used as input parameters for GMM. The experiment results for the dialect corpus of Vietnamese shows that the recognition rate is 66.3% without F0 information and this recognition rate increases to 72.2% with F0 information.