Noise-vocoding strips speech intelligibility by replacing fine spectral detail with modulated noise

Evidence

In order to alter the intelligibility of the speech, a method known as noise-vocoding was implemented (Shannon et al. 1995 ). This method filters the speech into a number of frequency-bands, and uses the amplitude envelope of each band to modulate band-limited noise. — Dijk, Pim van, Physiology, Psychoacoustics and…, 4376

Audio-book versions of two classic works of fiction read by the same American English speaker were partitioned into speech snippets, each with a duration of 10s. 120 snippets were randomly selected for the experiment. In order to alter the intelligibility of the speech, a method known as noise-vocoding was implemented (Shannon et al. 1995 ). This method filters the speech into a number of frequency-bands, and uses the amplitude envelope of each band to modulate band-limited noise. In our experiment, frequency-bands were logarithmically spaced between 70 and 5000 Hz. To determine the number of frequency-bands to use, an offline intelligibility test was performed for each subject using a separate set of 40 speech snippets. The number of frequency-bands was chosen such that subjects could understand an average of two words in each snippet. — Dijk, Pim van, Physiology, Psychoacoustics and…, 4380

Categorization

  • DEFINE: Explains what noise-vocoding is — a standard technique for degrading speech intelligibility while preserving its amplitude envelope.
  • XREF: Connects to the classic Shannon et al. 1995 vocoding literature establishing that spectral detail is dispensable once temporal envelopes are preserved.
  • corroborated: 2 independent passages support this claim — Dijk, Pim van, Physiology, Psychoacoustics and…, 4376; Dijk, Pim van, Physiology, Psychoacoustics and…, 4380