Creation of acoustic signal dictionary for ESNOLA based concatenated Bangla and Nepali TTS system

Present paper describes the detail design and development of two different acoustic signal dictionaries for incorporating separately into Epoch Synchronous Non OverLap Add (ESNOLA) method based Bangla (SCB) and Nepali concatenative ´ITS system. Present work uses a new set of signal units in sub-phonemic level, namely, Partnemes and allows a flexible approach to the length of transitions. Partnemes include identifiable portions unique for phonemes, their transitions and co-articulations. The creation process includes a series of normalization (Pitch, Amplitude and DC) with judicial selection and augmentation of speech segments such that smaller fundamental yet appropriate parts of the phonemes, interphoneme and inter-word transitions can be used as acoustic units. Encouraging results of the listening test confirm good perceptual quality and acceptability of the developed signal dictionaries. ESNOLA framework with optimal size partneme inventories altogether give a simple approach for generation of high quality synthesized speech with easy portability to hand-held devices.