A microscopic investigation of the effect of random envelope fluctuations on phoneme-in-noise perception

Autor: Alejandro Osses, Léo Varnet
Rok vydání: 2022
Předmět:
DOI: 10.17605/osf.io/ya6v7
Popis: This study will investigate the effect of specific noise realizations on the perception of speech acoustic cues in a consonant-in-noise categorization task using nonsense words of the structure vowel-consonant-vowel (VCV). More specifically, the discrimination will be tested using the utterances /aba/ and /ada/, measuring the discriminability of the middle-consonant phoneme presented in one of three types of background noises. All noises have the same flat long-term averaged spectrum but differ in the amount of spectro-temporal modulations: A Gaussian white noise, a noise with limited modulation spectrum (MPSL noise, see Study design), or a noise with randomly-imposed time-frequency regions of enhancement (bump noise, see Study design). The perceptual effect of noise is usually investigated using a macroscopic approach, by averaging the participant's responses over a number of target-plus-noise trials. Here, we will follow a microscopic approach by relating the participant's responses to the specific properties of the noise in each trial. This approach will be used to assess the random effect of noise, which is related to how a specific realization of the noise can bias the listeners' judgments towards one phoneme or the other. Our driving hypothesis is that the use of background noises with more random envelope fluctuations will result in a stronger random effect of the noise. These random envelope fluctuations are the result of the processing of incoming signals---here noisy maskers---at the output of each cochlear channel in the inner ear, that are further projected to higher stages of the auditory pathway. We assume that the cochlear channels have a width of one equivalent rectangular bandwidth (ERB) and that these fluctuations can be detrimental to our speech categorization task. We will comprehensively describe the statistical properties of the random background noises, before and after the cochlear processing, and will link these properties to the behavioral results of our listeners. The overall effect of noise on this phoneme-discrimination task will be assessed from the trial signal-to-noise (SNR) ratios using the standard psychophysical metrics of threshold and discrimination sensitivity. In line with previous work, the random effect of noise will be quantified using Auditory Classification Images (ACIs). The ACI is an hypothesis-neutral approach that measures the "listening strategy" of a participant during an auditory categorization task by detecting the time-frequency acoustic characteristics of the stimuli that are effectively used by the listener to solve the task (Varnet et al. 2013, 2015). As the objective of this study is to measure the random effect of noise, the ACIs will only be derived using the spectro-temporal configuration of the noise, while disregarding other important characteristics of the stimuli such as the spectro-temporal information and the presentation level of the target speech samples. While this methodological choice will likely yield less-than-optimal ACI predictions, it will allow us to use the ACI prediction performance as a proxy for the strength of the random effect of noise in a given condition for a given participant. Finally, the effect of noise on phoneme perception will be simulated using a perceptual model of the human auditory system in the same experimental conditions. This computational model uses primarily signal-driven (bottom-up) cues combined with a simple (top-down) decision back-end based on template matching (Osses & Kohlrausch, 2021). This model will be used as an "artificial listener," and will serve to confirm the validity of our methodological assumptions and metrics.
Databáze: OpenAIRE
Externí odkaz: