Speech Intelligibility for Target and Masker with Different Spectra.

Area of Science:

• Acoustics
• Psychoacoustics
• Speech Perception

Background:

• The Speech Intelligibility Index (SII) assumes signal-to-noise ratios (SNRs) outside [-15 dB; +15 dB] do not affect speech intelligibility.
• This assumption is crucial for predicting speech understanding in various acoustic environments.

Purpose of the Study:

• To experimentally evaluate the validity of the SII's assumed effective SNR range for speech intelligibility.
• To investigate how speech reception thresholds (SRTs) are affected by SNRs beyond the conventional [-15 dB; +15 dB] range using filtered signals.

Main Methods:

• Four experiments measured SRTs using speech and speech-spectrum noise.
• Target or noise signals were filtered (high-pass or low-pass) above/below 1400 Hz with varying attenuation levels.
• This manipulation created different SNRs in specific frequency bands to test SII assumptions.

Main Results:

• SRTs initially varied linearly with attenuation but reached an asymptote at high levels.
• The -15 dB SII limit held for high-pass filtered targets but not low-pass filtered targets.
• Speech intelligibility continued to improve with increasing SNR beyond +15 dB for filtered noise, up to +43 dB.
• Filtering noise had a greater impact on SRT reduction than filtering the target signal.

Conclusions:

• The study's findings challenge the SII's assumed SNR range and importance function, particularly for sharply filtered speech signals.
• The results suggest that the SII may overestimate or underestimate speech intelligibility in conditions with spectrally limited signals.
• Further research is needed to refine the SII model for more accurate predictions in complex auditory scenarios.