Integration of Phonotactic Features for Language Identification on Code-Switched Speech

In this paper, phoneme sequences are used as language information to perform code-switched identification (LID). With the one-pass recognition system, spoken sounds converted into phonetically arranged of sounds. The acoustic models robust enough handle multiple languages when emulating hidden Markov (HMMs). To determine similarity among our target languages, we reported two methods mapping. Statistical phoneme-based bigram (LM) integrated speech decoding eliminate possible phone mismatches. supervised support vector machine (SVM) is learn recognize phonetic mixed-language based on recognized sequences. As back-end decision taken by an SVM, likelihood scores segments with monolingual occurrence classify identity. corpus was tested Sepedi and English that often mixed. Our system evaluated measuring both ASR performance LID separately. systems have obtained a promising accuracy data-driven merging approach modelled using 16 Gaussian mixtures per state. respectively, proposed achieved acceptable accuracy.