The intelligibility of noise-vocoded speech:spectral information available from across-channel comparison of amplitude envelopes

Noise-vocoded (NV) speech is often regarded as conveying phonetic information primarily through temporal-envelope cues rather than spectral cues. However, listeners may infer the formant frequencies in the vocal-tract output—a key source of phonetic detail—from across-band differences in amplitude when speech is processed through a small number of channels. The potential utility of this spectral information was assessed for NV speech created by filtering sentences into six frequency bands, and using the amplitude envelope of each band (=30 Hz) to modulate a matched noise-band carrier (N). Bands were paired, corresponding to F1 (˜N1 + N2), F2 (˜N3 + N4) and the higher formants (F3' ˜ N5 + N6), such that the frequency contour of each formant was implied by variations in relative amplitude between bands within the corresponding pair. Three-formant analogues (F0 = 150 Hz) of the NV stimuli were synthesized using frame-by-frame reconstruction of the frequency and amplitude of each formant. These analogues were less intelligible than the NV stimuli or analogues created using contours extracted from spectrograms of the original sentences, but more intelligible than when the frequency contours were replaced with constant (mean) values. Across-band comparisons of amplitude envelopes in NV speech can provide phonetically important information about the frequency contours of the underlying formants.

The effect of dialect mismatch on likelihood-maximising speech enhancement for noise-robust speech recognition

Traditional speech enhancement methods optimise signal-level criteria such as signal-to-noise ratio, but these approaches are sub-optimal for noise-robust speech recognition. Likelihood-maximising (LIMA) frameworks are an alternative that optimise parameters of enhancement algorithms based on state sequences generated for utterances with known transcriptions. Previous reports of LIMA frameworks have shown significant promise for improving speech recognition accuracies under additive background noise for a range of speech enhancement techniques. In this paper we discuss the drawbacks of the LIMA approach when multiple layers of acoustic mismatch are present – namely background noise and speaker accent. Experimentation using LIMA-based Mel-filterbank noise subtraction on American and Australian English in-car speech databases supports this discussion, demonstrating that inferior speech recognition performance occurs when a second layer of mismatch is seen during evaluation.