You are here

Deep Learning for ASR

Investigate the use of deep learning approaches for speech recognition, speaker verification/recognition and speech analytics.


This project, funded by the Italian company PerVoice (, is two fold

  • the development of ASR systems for transcribing TV news in: Russian, Chinese and Japanese languages.
  • the development of a speaker diarization system based on the usage of DNN embedded representations of speaker identities.

Speech recognition

The usage of deep neural networks in combination with hidden Markov models (DNN-HMMs) has achieved great success in a number of works
dealing with many applications: mobile voice search, broadcast news transcription, automatic video subtitling, conversational speech recognition, even in noisy environments.