Spoken language interfaces (2)
Synthetic voice can be produced by the system MBROLA developed at the Faculte Polytechnique de Mons, Belgium. MBROLA is based on diphone concatenation techniques and is used to generate speech from a phonetic transcription.
Improvements in speech recognition have been made towards three directions:
- from isolated to continuous speech,
- from speaker-dependent to speaker independent, and,
- the increase in the vocabulary size.
Driving force: the use of Hidden Markov Models (HMMs).