The Human ear can easily hears natural words in complex noise-environment. Even this can hear some wanted words in bubble-environment. So we wish that noise-robust feature extraction through analysis about structure of human-ear.
The structure of auditory model consist of band-pass filters having the character of basilar membrane and hair-cell model. This paper uses Auditory Image Model(AIM) which have 20-channels time-domain filterbank which center frequencys are from 64Hz to 3500Hz.
And then, AIM has Temporal Integration stage modeling Inferior Colliculus after hair cell stage. AIM is expected as being able to extract better some features and suggest method of noise-robust feature extraction.
