 |
Workshop on the Auditory Basis of Speech PerceptionKeele University, UK |
 |
|
Workshop on the Auditory Basis of Speech PerceptionKeele University, UK |
|
A binaural auditory model, known as the Cocktail-Party-Processor, was used as a front-end in isolated word recognition based on a HMM recognizer. Experiments which compared the performance of the binaural models' representation to that of a standard monaural one showed that the binaural model per- formed significantly better in terms of recognition score for different kinds of additive noise (pink noise, concurrent speech) and a wide range of signal-to-noise ratios. Since the binaural representation is robust towards variations of the direction of incidence of the extracted signal, a new training of the speech recognizer for different acoustic situations is not necessary.
Bibliographic reference. Bodden, Markus / Rateitschek, Klaus (1996): "Noise-robust speech recognition based on a binaural auditory model", In ABSP-1996, 291-296.