Infants’ discrimination of spectral slope

Spectral slope is a global property of the spectral envelope representing the change in energy across spectral frequency. Spectral slope is a dimension for discriminating speech, voices and musical instrument timbres. The average spectral slope of speech and music is about −6 dB/octave. If spectral slope is important in timbre perception, enhanced sensitivity for spectral slopes around −6 dB/octave would be expected, and poor discrimination would be expected for spectral slopes very different from this value. Eight-month-old infants were tested across a range of spectral slopes (−16 to +16 dB/octave) to determine whether infants’ slope discrimination is best near −6 dB/octave. The stimuli were generated according to the following equation: dBi=b<th>log2<th>i, where i is the harmonic number, b is the spectral slope value, and dBi is the relative amplitude of harmonic i. The results show that infants were only sensitive to differences in spectral slope in a limited range between −10 and −4 dB/octave, despite the fact that local intensity cues were smallest in this range. This suggests that infants are using spectral slope as a dimension for discriminating timbre and that the auditory system is tuned to sounds with real-world spectral slopes. [Research supported by NSERC.]

Infants’ discrimination of spectral slope Journal Articles