14th International Congress of Phonetic Sciences (ICPhS-14)
San Francisco, CA, USA
Letter-to-phone conversion, as part of the natural language processing stage, plays a very important role in text-to-speech (TTS) synthesis because it associates an appropriate phonetic transcription with each word of the sentence to be pronounced. The classical approach for the phonetic conversion is based in most TTS systems on either a dictionary or a set of rules. Because both methods have important drawbacks, we used a completely different approach: we designed a very accurate grapheme-to-allophone converter for unrestricted TTS synthesis in Romanian, based on a parallel architecture of neural networks. The percentage of correctly transcribed words is extremely high, comparatively with both classical methods and other related implementations. The paper motivates the neural approach, describes the system architecture, gives implementation details and discusses the main results obtained for Romanian language.
Bibliographic reference. Burileanu, Dragoş / Sima, Mihai / Neagu, Adrian (1999): "A phonetic converter for speech synthesis in Romanian", In ICPhS-14, 503-506.