Rule-based Approach for English-Indonesian Code-switching Acoustic Model

The usage of English in daily conversation and many aspects in everyday life is increasing in Indonesia. It affects the usage of Indonesian as the main language. People often communicate using both English and Indonesian (code-switching). English-Indonesian code-switching causes degradation of the performance of Indonesian automatic speech recognition system. Other than out of vocabulary of foreign words, the system fails to recognize the foreign phonemes and predicts them as the most similar Indonesian phonemes. An investigation is conducted to improve the Indonesian automatic speech recognition system. To support the investigation, an English-Indonesian code-switching corpus is built. Rule-based approaches for English-Indonesian code-switching acoustic model are developed and experimented. Those approaches are English to Indonesian phone mapping and International Phonetic Alphabets based phone merging. The phoneme mapping approach succeeds to decrease the word error rate from the baseline system. This method outperformed the phone merging approach.