15th Annual Conference of the International Speech Communication Association

September 14-18, 2014

Principal Components of Auditory Spectro-Temporal Receptive Fields

Nagaraj Mahajan (1), Nima Mesgarani (2), Hynek Hermansky (1)

(1) Johns Hopkins University, USA
(2) Columbia University, USA

More than two thousand auditory cortical spectro-temporal receptive fields (STRFs) of the ferret were analysed by Principal Component Analysis (PCA) to reveal their dominant properties. Results show that cortical levels of mammalian auditory processing enhance relatively low modulation spectral components of the signal around 3 Hz, using relatively broad spectral processing channels of the order of octave or more, evaluating mean spectral values and local spectral slopes within such frequency channels. These observations are consistent with some engineering techniques shown to be effective in machine recognition of speech.

Full Paper

Bibliographic reference.  Mahajan, Nagaraj / Mesgarani, Nima / Hermansky, Hynek (2014): "Principal components of auditory spectro-temporal receptive fields", In INTERSPEECH-2014, 1983-1987.