Chimera Synthesis
Home Why Study Chimeras? Chimera Synthesis Chimera Demos Further Information

 

                                                                                   

 
bulletTwo sounds (e.g. speech and music) are used as input. Each sound is split into 1-64 complementary frequency bands spanning the range 80-4400 Hz with a perfect-reconstruction filter bank. The filters are nearly rectangular, and their center frequencies are equally spaced along the human cochlear frequency map (Greenwood, 1990). 
bulletFiltered signals from matching frequency bands are processed through a chimerizer, which exchanges the envelope and the fine structure of the two signals, producing two partial chimeras A and B.
bulletPartial chimeras of the same type (A or B), are summed over all frequency bands to produce two multi-band chimeras A and B.  Chimera A has, in each frequency band, the envelope of Sound 1 and the fine structure of Sound 2.  Chimera B has the envelope of the Sound 2 and the fine structure of Sound 1.

Chimerizer

bulletEach band-limited input signal is factored into its envelope and fine structure using the Hilbert transform. This expresses each signal s(t) as the product a(t) cos f(t), where a(t) is the envelope and cos f(t) the fine structure.
bulletThe partial chimeras cA(t) and cB(t) are synthesized by multiplying the envelope of one filtered signal s1(t) by the fine structure of the other signal s2(t): cA(t) = a1(t) cos f2(t), and cB(t)= a2(t) cos f1(t).
Greenwood, D.D. A cochlear frequency-position function for several species - 29 years later. J. Acoust. Soc.  Am. 87: 2592-2604, 1990
 

Chimera Spectrograms

 

bullet

Chimeras with 1, 4 and 16 bands were synthesized from (1) the utterance "Wood is best for making toys and blocks", and (2) noise have the same long-term power spectrum.

bullet

Broadband spectrograms of 1-band chimeras resemble neither of the original spectrograms. However, speech formants can partly be seen in the spectrogram for the chimera having the speech fine structure, and the coarse temporal structure of speech is obvious in the chimera with speech envelope.

bullet

As the number of bands increases, spectrograms of chimeras increasingly resemble the spectrogram of the original sound which has the same envelope.

These results show that envelope information increasingly dominates in the spectrogram as the number of frequency bands increases, and therefore the width of each band decreases.