Index of content:
Volume 126, Issue 3, September 2009
- SPEECH PROCESSING AND COMMUNICATION SYSTEMS 
126(2009); http://dx.doi.org/10.1121/1.3184603View Description Hide Description
Traditional noise-suppression algorithms have been shown to improve speech quality, but not speech intelligibility. Motivated by prior intelligibility studies of speech synthesized using the ideal binary mask, an algorithm is proposed that decomposes the input signal into time-frequency (T-F) units and makes binary decisions, based on a Bayesian classifier, as to whether each T-F unit is dominated by the target or the masker. Speech corrupted at low signal-to-noise ratio (SNR) levels ( and ) using different types of maskers is synthesized by this algorithm and presented to normal-hearing listeners for identification. Results indicated substantial improvements in intelligibility (over 60% points in babble) over that attained by human listeners with unprocessed stimuli. The findings from this study suggest that algorithms that can estimate reliably the SNR in each T-F unit can improve speech intelligibility.
126(2009); http://dx.doi.org/10.1121/1.3179668View Description Hide Description
In the presence of noise, do speakers actively shift their spectral energy distribution to regions least affected by the noise? The current study measured speech level, fundamental frequency, first formant frequency, and spectral center of gravity for read speech produced in the presence of low- and high-pass filtered noise. In both filtering conditions, these acoustic parameters increased relative to speech produced in quiet, a response which creates a release from masking for listeners in the low-pass condition but which actually increases masking in the high-pass noise condition. These results suggest that, at least for read speech, speakers do not adopt production strategies in noise which optimize listeners’ information reception but that instead the observed shifts could be a passive response which creates a fortuitous masking release in the low-pass noise. Independent variation in parameters such as F0, F1 and spectral center of gravity may be severely constrained by the increase in vocal effort which accompanies Lombard speech.
126(2009); http://dx.doi.org/10.1121/1.3183593View Description Hide Description
Differences in speaking style are associated with more or less spectral variability, as well as different modulation characteristics. The greater variation in some styles (e.g., spontaneous speech and infant-directed speech) poses challenges for recognition but possibly also opportunities for learning more robust models, as evidenced by prior work and motivated by child language acquisition studies. In order to investigate this possibility, this work proposes a new method for characterizing speaking style (the modulation spectrum), examines spontaneous, read, adult-directed, and infant-directed styles in this space, and conducts pilot experiments in style detection and sampling for improved speech recognizer training. Speaking style classification is improved by using the modulation spectrum in combination with standard pitch and energy variation. Speech recognition experiments on a small vocabulary conversational speech recognition task show that sampling methods for training with a small amount of data benefit from the new features.