The Idiap Research Institute seeks qualified candidates for one PhD position in the area of acoustic data driven grapheme-to-subword unit conversion applied to automatic speech recognition (ASR).
In standard hidden Markov (HMM) based ASR systems or text-to-speech (TTS) systems, each word is modelled as a sequence of subword units. The subword units typically being phonemes/phones. The sequence of subword units is often referred to as pronunciation model of the word, e.g. CAT - /k/ /ae/ /t/. The lexicon of the ASR or TTS system contains the mapping between the word and its pronunciation model.
More info about the position can be found on our "".