Channel selection in the modulation domain for improved speech intelligibility in noise

Area of Science:

• Auditory Neuroscience
• Speech Processing
• Signal Processing

Background:

• Background noise significantly degrades speech intelligibility by masking crucial low-frequency envelope modulations.
• The modulation signal-to-noise ratio ((S/N)(mod)) quantifies the relative strength of target versus masker envelope modulations.

Purpose of the Study:

• To investigate the impact of modulation-selective processing on speech intelligibility in noise.
• To determine if preserving target-dominant modulations and discarding masker-dominant ones improves speech reception thresholds.

Main Methods:

• Speech envelopes were decomposed into modulation frequency channels (0-30 Hz).
• A (S/N)(mod) criterion was used to identify and retain target-dominant modulations.
• Masker-dominant modulations were discarded from the mixture envelopes.

Main Results:

• Processing that preserved target-dominant modulations improved speech reception thresholds by up to 13 dB.
• This benefit was observed for modulations up to 18 Hz.
• Discarding masker-dominant modulations was key to the observed intelligibility improvement.

Conclusions:

• Selective processing of speech envelope modulations can significantly enhance intelligibility in noisy conditions.
• The (S/N)(mod) measure provides an effective criterion for identifying perceptually relevant modulations.
• This approach offers a promising strategy for noise-reduction algorithms aimed at improving speech perception.