Developing a basque TTS for the navarro-lapurdian dialect

Abstract

The paper presents a new TTS system for the Navarro- Lapurdian dialect based on a standard Basque TTS. A phonetically balanced recording corpus of 4000 sentences has been designed and two speakers have recorded it. The voice has been built using a high quality speech coder in the context of HMM based speech synthesis. The new dialectal TTS system has been compared in a subjective evaluation with the existing TTS system for standard Basque and with a mixed system that applies the phonetic transcription rules of the dialect, but uses the speech generation module of the standard Basque system. The adaptation of the front-end module with the inclusion of new phonetic transcription rules and new sounds is not enough to get a system that works better than the standard Basque system. The results with the dialectal new voice indicate that users prefer the new dialectal system to the standard Basque one.