Hehrmann, Phillipp; (2018) Pitch perception as probabilistic inference. Doctoral thesis (Ph.D), UCL (University College London).

Preview

Text Hehrmann_10062222_thesis.pdf Download (7MB) | Preview

Abstract

Pitch is a fundamental and salient perceptual attribute of many behaviourally important sounds, including animal calls, human speech and music. Human listeners perceive pitch without conscious effort or attention. These and similar observations have prompted a search for mappings from acoustic stimulus to percept that can be easily computed from peripheral neural responses at early stages of the central auditory pathway. This tenet however is not supported by physiological evidence: how the percept of pitch is encoded in neural firing patterns across the brain, and where – if at all – such a representation may be localised remain as yet unsolved questions. Here, instead of seeking an explanation guided by putative mechanisms, we take a more abstract stance in developing a model by asking, what computational goal the auditory system is set up to achieve during pitch perception. Many natural pitch-evoking sounds are approximately periodic within short observation time windows. We posit that pitch reflects a near-optimal estimate of the underlying periodicity of sounds from noisy evoked responses in the auditory nerve, exploiting statistical knowledge about the regularities and irregularities occurring during sound generation and transduction. We compute (or approximate) the statistically optimal estimate using a Bayesian probabilistic framework. Model predictions match the pitch reported by human listeners for a wide range of welldocumented, pitch-evoking stimuli, both periodic and aperiodic. We then present new psychophysical data on octave biases and pitch-timbre interactions in human perception which further demonstrates the validity of our approach, while posing difficulties for alternative models based on autocorrelation analysis or simple spectral pattern matching. Our model embodies the concept of perception as unconscious inference, originally proposed by von Helmholtz as an interface bridging optics and vision. Our results support the view that even apparently primitive acoustic percepts may derive from subtle statistical inference, suggesting that such inferential processes operate at all levels across our sensory systems.

Download activity - last month

Export as XMLCSVJSON

Download activity - last 12 months

Export as JSONXMLCSV

Downloads by country - last 12 months

Export as XMLCSVJSON

Archive Staff Only

View Item