Cross-lingual phoneme mapping for multilingual synthesis systems.

Development of a multilingual text-to-speech (TTS) system usually requires a lot of time, effort and language resources. Furthermore, the implementation tends to consume large amounts of memory as the number of supported languages increases. This paper proposes a simple method for quickly increasing the language portfolio of an existing TTS system with the minimal effort and memory consumption. The crosslingual phoneme mapping modifies the phonetic transcription of a new language by presenting it with the phoneme set supported by an existing TTS system. The synthesis output of the mapped language is phonetically quite accurate but the intonation is based on the existing target language. The evaluations showed that the cross-lingual phoneme mapping could provide an adequate quality synthesis when the language portfolio must rapidly be expanded and the memory and the language resources are scarce. The method is applied best for the synthesis of short utterances including names and isolated words.