Defining spectral and temporal resolutions of information-bearing acoustic changes for understanding noise-vocoded sentences

Abstract

Information-bearing acoustic changes (IBACs) in the speech signal are highly important for speech perception. This has been demonstrated for full-spectrum sentences using cochlea-scaled entropy (CSE; Stilp and Kluender, 2010 PNAS) and noise-vocoded sentences using an adapted metric (CSECI; Stilp et al., 2013 JASA). While IBACs appear fundamental to speech perception most broadly, Stilp et al. tested a single set of vocoder parameters, obscuring the breadth and depth of perceptual reliance upon these acoustic changes. Here, TIMIT sentences were noise-vocoded with variable spectral resolution (4-24 spectral channels spanning 300–5000 Hz), variable temporal resolution (4–64 Hz amplitude envelope cutoff frequency), or combinations therein. High-CSECI or low-CSECI sentence intervals were replaced with speech-shaped noise. As spectral resolution decreased, IBACs became more important for sentence understanding, especially at 6–12 channels. At high spectral resolutions, performance nearly overcame replacement of low-CSECI intervals but not high-CSECI intervals. Importance of IBACs at different temporal resolutions was largely driven by overall intelligibility, suggesting CSECI has sufficient temporal resolution at low modulation frequencies. Data exploring spectral-temporal tradeoffs will also be presented. Peak-picking strategies in CIs select and stimulate channels according to their amplitudes, but results suggest additional perceptual benefit may be offered by encoding IBACs as well.