Intelligibility of 1/3-octave speech: greater contribution of frequencies outside than inside the nominal passband

We reported previously that "everyday" sentences were highly intelligible when limited to a 1/3-octave passband centered at 1,500 Hz and having transition-band slopes of approximately 100 dB/octave. The present study determined the relative contributions to intelligibility made by the passband (PB) and the transition bands (TBs) by partitioning the same bandpass sentences using 2,000-order FIR filtering. Intelligibility scores were: PB with both TBs, 92%; deletion of both TBs (leaving only the 1/3-octave PB with nearly vertical slopes), 24%; deletion of the PB (leaving both TBs separated by a 1/3-octave gap), 83%. These and other results indicate a remarkable ability to compensate for severe spectral tilt and the consequent importance of considering frequencies outside the nominal passband in interpreting studies using filtered speech.     

Intelligibility of bandpass filtered speech: steepness of slopes required to eliminate transition band contributions

Despite the recognition that the steepness of filter slopes can play an important role in the intelligibility of bandpass speech, there has been no systematic examination of its importance. The present study used high orders of finite impulse response (FIR) filtering to produce slopes ranging from 150 to 10,000 dB/octave. The slopes flanked 1/3-octave passbands of everyday sentences having a center frequency of 1500 Hz (the region of highest intelligibility for the male speaker's voice). Presentation levels were approximately 75 and 45 dB. No significant differences were found for the two presentation levels. Average intelligibility scores ranged from 77% at 150 dB/octave down to the asymptotic intelligibility score of 12% at 4800 dB/octave. These results indicate that slopes of several thousand dB/octave may be required for accurate and unambiguous specification of the range of frequencies contributing to intelligibility of filtered speech. In addition, the extremely steep slopes are needed to ensure that none of the spectral components contributing to intelligibility has its relative importance diminished by spectral tilt.     

Syllable intelligibility for temporally filtered LPC cepstral trajectories.

The intelligibility of syllables whose cepstral trajectories were temporally filtered was measured. The speech signals were transformed to their LPC cepstral coefficients, and these coefficients were passed through different filters. These filtered trajectories were recombined with the residuals and the speech signal reconstructed. The intelligibility of the reconstructed speech segments was then measured in two perceptual experiments for Japanese syllables. The effect of various low-pass, high-pass, and bandpass filtering is reported, and the results summarized using a theoretical approach based on the independence of the contributions in different modulation bands. The overall results suggest that speech intelligibility is not severely impaired as long as the filtered spectral components have a rate of change between 1 and 16 Hz.     

The ability of listeners to use recovered envelope cues from speech fine structure

Recent work has demonstrated that auditory filters recover temporal-envelope cues from speech fine structure when the former were removed by filtering or distortion. This study extended this work by assessing the contribution of recovered envelope cues to consonant perception as a function of the analysis bandwidth, when vowel-consonant-vowel (VCV) stimuli were processed in order to keep their fine structure only. The envelopes of these stimuli were extracted at the output of a bank of auditory filters and applied to pure tones whose frequency corresponded to the original filters' center frequencies. The resulting stimuli were found to be intelligible when the envelope was extracted from a single, wide analysis band. However, intelligibility decreases from one to eight bands with no further decrease beyond this value, indicating that the recovered envelope cues did not play a major role in consonant perception when the analysis bandwidth was narrower than four times the bandwidth of a normal auditory filter (i.e., number of analysis bands > or =8 for frequencies spanning 80 to 8020 Hz).     

单通道语音增强算法对汉语语音可懂度影响的研究

考察了当前常用的几种单通道语音增强算法对汉语语音可懂度的影响。受不同类型噪音干扰的语音经过5种单通道语音增强算法的处理后,播放给具有正常听力水平的被试进行听辩,考察增强后语音的可懂度。实验结果表明,语音增强算法并不能改进语音的可懂度水平;通过分析具体的错误原因,发现听辩错误主要来自于音素错误,与声调关系不大;而且,同英文的辨识结果相比,一些增强算法对于中、英文可懂度影响差异显著。      

汉语语音合成系统评价方法

从1994年开始,对汉语语音合成系统的工作性能定期举行全国评测.采用语言清晰度测试方法,1994年对五个不同的合成系统进行了评测和诊断.听音人为16名大学生(男8,女8),对合成言语没有经验.听音人响应是开放的听音记录.同时,还采用十点主观评价(MOS)测定言语自然度.为给出各合成系统音段层的诊断信息,对合成语音的辅音知觉混淆矩阵进行了分析.借助于对比自然言语和合成言语在不同语言层次上清晰度试验得分间的统计关系,来考察合成系统韵律特征处理的缺陷.结果表明,采用上述方法可得到评测合成系统工作性能的稳定合理的指标.有关韵律特征的评价方法有待于进一步发展.     

Contribution of frequency modulation to speech recognition in noise

Cochlear implants allow most patients with profound deafness to successfully communicate under optimal listening conditions. However, the amplitude modulation (AM) information provided by most implants is not sufficient for speech recognition in realistic settings where noise is typically present. This study added slowly varying frequency modulation (FM) to the existing algorithm of an implant simulation and used competing sentences to evaluate FM contributions to speech recognition in noise. Potential FM advantage was evaluated as a function of the number of spectral bands, FM depth, FM rate, and FM band distribution. Barring floor and ceiling effects, significant improvement was observed for all bands from 1 to 32 with the additional FM cue both in quiet and noise. Performance also improved with greater FM depth and rate, which might reflect resolved sidebands under the FM condition. Having FM present in low-frequency bands was more beneficial than in high-frequency bands, and only half of the bands required the presence of FM, regardless of position, to achieve performance similar to when all bands had the FM cue. These results provide insight into the relative contributions of AM and FM to speech communication and the potential advantage of incorporating FM for cochlear implant signal processing.     

Acoustic and linguistic factors in the perception of bandpass-filtered speech

Speech can remain intelligible for listeners with normal hearing when processed by narrow bandpass filters that transmit only a small fraction of the audible spectrum. Two experiments investigated the basis for the high intelligibility of narrowband speech. Experiment 1 confirmed reports that everyday English sentences can be recognized accurately (82%-98% words correct) when filtered at center frequencies of 1500, 2100, and 3000 Hz. However, narrowband low predictability (LP) sentences were less accurately recognized than high predictability (HP) sentences (20% lower scores), and excised narrowband words were even less intelligible than LP sentences (a further 23% drop). While experiment 1 revealed similar levels of performance for narrowband and broadband sentences at conversational speech levels, experiment 2 showed that speech reception thresholds were substantially (>30 dB) poorer for narrowband sentences. One explanation for this increased disparity between narrowband and broadband speech at threshold (compared to conversational speech levels) is that spectral components in the sloping transition bands of the filters provide important cues for the recognition of narrowband speech, but these components become inaudible as the signal level is reduced. Experiment 2 also showed that performance was degraded by the introduction of a speech masker (a single competing talker). The elevation in threshold was similar for narrowband and broadband speech (11 dB, on average), but because the narrowband sentences required considerably higher sound levels to reach their thresholds in quiet compared to broadband sentences, their target-to-masker ratios were very different (+23 dB for narrowband sentences and -12 dB for broadband sentences). As in experiment 1, performance was better for HP than LP sentences. The LP-HP difference was larger for narrowband than broadband sentences, suggesting that context provides greater benefits when speech is distorted by narrow bandpass filtering.     

Sentential, lexical, and acoustic effects on the perception of word boundaries

This study investigates the effects of sentential context, lexical knowledge, and acoustic cues on the segmentation of connected speech. Listeners heard near-homophonous phrases (e.g., plmpaI for "plum pie" versus "plump eye") in isolation, in a sentential context, or in a lexically biasing context. The sentential context and the acoustic cues were piloted to provide strong versus mild support for one segmentation alternative (plum pie) or the other (plump eye). The lexically biasing context favored one segmentation or the other (e.g., skmpaI for "scum pie" versus *"scump eye," and lmpaI, for "lump eye" versus *"lum pie," with the asterisk denoting a lexically unacceptable parse). A forced-choice task, in which listeners indicated which of two words they thought they heard (e.g., "pie" or "eye"), revealed compensatory mechanisms between the sources of information. The effect of both sentential and lexical contexts on segmentation responses was larger when the acoustic cues were mild than when they were strong. Moreover, lexical effects were accompanied with a reduction in sensitivity to the acoustic cues. Sentential context only affected the listeners' response criterion. The results highlight the graded, interactive, and flexible nature of multicue segmentation, as well as functional differences between sentential and lexical contributions to this process.     

Laminar cortical dynamics of conscious speech perception: neural model of phonemic restoration using subsequent context in noise

How are laminar circuits of neocortex organized to generate conscious speech and language percepts? How does the brain restore information that is occluded by noise, or absent from an acoustic signal, by integrating contextual information over many milliseconds to disambiguate noise-occluded acoustical signals? How are speech and language heard in the correct temporal order, despite the influence of contexts that may occur many milliseconds before or after each perceived word? A neural model describes key mechanisms in forming conscious speech percepts, and quantitatively simulates a critical example of contextual disambiguation of speech and language; namely, phonemic restoration. Here, a phoneme deleted from a speech stream is perceptually restored when it is replaced by broadband noise, even when the disambiguating context occurs after the phoneme was presented. The model describes how the laminar circuits within a hierarchy of cortical processing stages may interact to generate a conscious speech percept that is embodied by a resonant wave of activation that occurs between acoustic features, acoustic item chunks, and list chunks. Chunk-mediated gating allows speech to be heard in the correct temporal order, even when what is heard depends upon future context.     

A new efficient two-channel backward algorithm for speech intelligibility enhancement: A subband approach

This paper addresses the problem of speech intelligibility enhancement by adaptive filtering algorithms employed with subband techniques. The two structures named the forward and backward blind source separation structures are extensively used in the speech enhancement and source separation areas, and largely studied in the literature with convolutive and non-convolutive mixtures. These two structures use two-microphones to generate the convolutive/non-convolutive mixing signal, and provide at the outputs the target and the jammer signal components. In this paper, we focus our interest on the backward structure employed to enhance the speech signal from a convolutive mixture. Furthermore, we propose a subband implementation of this structure to improve its behavior with speech signal. The new proposed subband-Backward BSS (SBBSS) structure allows a very important improvement of the convergence speed of the adaptive filtering algorithms when the subband-number is selected high. In order to improve the robustness of the proposed subband structure, we have adapted then applied a new criterion that combines the System Mismatch and the Mean-Errors criterion minimization. The proposed subband backward structure, when it is combined with this new criterion minimization, allows to enhance the output speech signal by reducing the distortion and the noise components. The performance of the proposed subband backward structure is validated through several objective criteria which are given and described in this paper.     

Auditory versus phonetic accounts of observed confusions between consonant phonemes.

The utility of phonetic features versus acoustic properties for describing perceptual relations among speech sounds was evaluated with a multidimensional scaling analysis of Miller and Nicely's [J. Acoust. Soc. Am. 27, 338-352 (1955)] consonant confusions data. The INDSCAL method and program were employed with the original data log transformed to enhance consistency with the linear INDSCAL model. A four-dimensional solution accounted for 69% of the variance and was best characterized in terms of acoustic properties of the speech signal, viz., temporal relationship of periodicity and burst onset, shape of voiced first formanant transition, shape of voiced second formanant transition, and amount of initial spectral dispersion, rather than in terms of phonetic features. The amplitude and spectral location of acoustic energy specifying each perceptual dimension were found to determine a dimension's perceptual effect as the signal was degraded by masking noise and bandpass filtering. Consequently, the perceptual bases of identification confusions between pairs of syllables were characterized in terms of the shared acoustic properties which remained salient in the degraded speech. Implications of these findings for feature-based accounts of perceptual relationships between phonemes are considered.     

Spectral weighting strategies for sentences measured by a correlational method

Listeners' ability to understand speech in adverse listening conditions is partially due to the redundant nature of speech. Natural redundancies are often lost or altered when speech is filtered, such as done in AI/SII experiments. It is important to study how listeners recognize speech when the speech signal is unfiltered and the entire broadband spectrum is present. A correlational method [R. A. Lutfi, J. Acoust. Soc. Am. 97, 1333-1334 (1995); V. M. Richards and S. Zhu, J. Acoust. Soc. Am. 95, 423-424 (1994)] has been used to determine how listeners use spectral cues to perceive nonsense syllables when the full speech spectrum is present [K. A. Doherty and C. W. Turner, J. Acoust. Soc. Am. 100, 3769-3773 (1996); C. W. Turner et al., J. Acoust. Soc. Am. 104, 1580-1585 (1998)]. The experiments in this study measured spectral-weighting strategies for more naturally occurring speech stimuli, specifically sentences, using a correlational method for normal-hearing listeners. Results indicate that listeners placed the greatest weight on spectral information within bands 2 and 5 (562-1113 and 2807-11,000 Hz), respectively. Spectral-weighting strategies for sentences were also compared to weighting strategies for nonsense syllables measured in a previous study (C. W. Turner et al., 1998). Spectral-weighting strategies for sentences were different from those reported for nonsense syllables.     

Analysis of a simplified normalized covariance measure based on binary weighting functions for predicting the intelligibility of noise-suppressed speech

The normalized covariance measure (NCM) has been shown previously to predict reliably the intelligibility of noise-suppressed speech containing non-linear distortions. This study analyzes a simplified NCM measure that requires only a small number of bands (not necessarily contiguous) and uses simple binary (1 or 0) weighting functions. The rationale behind the use of a small number of bands is to account for the fact that the spectral information contained in contiguous or nearby bands is correlated and redundant. The modified NCM measure was evaluated with speech intelligibility scores obtained by normal-hearing listeners in 72 noisy conditions involving noise-suppressed speech corrupted by four different types of maskers (car, babble, train, and street interferences). High correlation (r = 0.8) was obtained with the modified NCM measure even when only one band was used. Further analysis revealed a masker-specific pattern of correlations when only one band was used, and bands with low correlation signified the corresponding envelopes that have been severely distorted by the noise-suppression algorithm and/or the masker. Correlation improved to r = 0.84 when only two disjoint bands (centered at 325 and 1874 Hz) were used. Even further improvements in correlation (r = 0.85) were obtained when three or four lower-frequency (<700 Hz) bands were selected.     

Recognition of spectrally degraded and frequency-shifted vowels in acoustic and electric hearing

The present study measured the recognition of spectrally degraded and frequency-shifted vowels in both acoustic and electric hearing. Vowel stimuli were passed through 4, 8, or 16 bandpass filters and the temporal envelopes from each filter band were extracted by half-wave rectification and low-pass filtering. The temporal envelopes were used to modulate noise bands which were shifted in frequency relative to the corresponding analysis filters. This manipulation not only degraded the spectral information by discarding within-band spectral detail, but also shifted the tonotopic representation of spectral envelope information. Results from five normal-hearing subjects showed that vowel recognition was sensitive to both spectral resolution and frequency shifting. The effect of a frequency shift did not interact with spectral resolution, suggesting that spectral resolution and spectral shifting are orthogonal in terms of intelligibility. High vowel recognition scores were observed for as few as four bands. Regardless of the number of bands, no significant performance drop was observed for tonotopic shifts equivalent to 3 mm along the basilar membrane, that is, for frequency shifts of 40%-60%. Similar results were obtained from five cochlear implant listeners, when electrode locations were fixed and the spectral location of the analysis filters was shifted. Changes in recognition performance in electrical and acoustic hearing were similar in terms of the relative location of electrodes rather than the absolute location of electrodes, indicating that cochlear implant users may at least partly accommodate to the new patterns of speech sounds after long-time exposure to their normal speech processor.     

Measuring the critical band for speech

The current experiments were designed to measure the frequency resolution employed by listeners during the perception of everyday sentences. Speech bands having nearly vertical filter slopes and narrow bandwidths were sharply partitioned into various numbers of equal log- or ERBN-width subbands. The temporal envelope from each partition was used to amplitude modulate a corresponding band of low-noise noise, and the modulated carriers were combined and presented to normal-hearing listeners. Intelligibility increased and reached asymptote as the number of partitions increased. In the mid- and high-frequency regions of the speech spectrum, the partition bandwidth corresponding to asymptotic performance matched current estimates of psychophysical tuning across a number of conditions. These results indicate that, in these regions, the critical band for speech matches the critical band measured using traditional psychoacoustic methods and nonspeech stimuli. However, in the low-frequency region, partition bandwidths at asymptote were somewhat narrower than would be predicted based upon psychophysical tuning. It is concluded that, overall, current estimates of psychophysical tuning represent reasonably well the ability of listeners to extract spectral detail from running speech.     

Speech disruption during delayed auditory feedback with simultaneous visual feedback

Delayed auditory feedback (DAF) regarding speech can cause dysfluency. The purpose of this study was to explore whether providing visual feedback in addition to DAF would ameliorate speech disruption. Speakers repeated sentences and heard their auditory feedback delayed with and without simultaneous visual feedback. DAF led to increased sentence durations and an increased number of speech disruptions. Although visual feedback did not reduce DAF effects on duration, a promising but nonsignificant trend was observed for fewer speech disruptions when visual feedback was provided. This trend was significant in speakers who were overall less affected by DAF. The results suggest the possibility that speakers strategically use alternative sources of feedback.     

Auralization study of optimum reverberation times for speech intelligibility for normal and hearing-impaired listeners in classrooms with diffuse sound fields

Speech-intelligibility tests auralized in a virtual classroom were used to investigate the optimal reverberation times for verbal communication for normal-hearing and hearing-impaired adults. The idealized classroom had simple geometry, uniform surface absorption, and an approximately diffuse sound field. It contained a speech source, a listener at a receiver position, and a noise source located at one of two positions. The relative output levels of the speech and noise sources were varied, along with the surface absorption and the corresponding reverberation time. The binaural impulse responses of the speech and noise sources in each classroom configuration were convolved with Modified Rhyme Test (MRT) and babble-noise signals. The resulting signals were presented to normal-hearing and hearing-impaired adult subjects to identify the configurations that gave the highest speech intelligibilities for the two groups. For both subject groups, when the speech source was closer to the listener than the noise source, the optimal reverberation time was zero. When the noise source was closer to the listener than the speech source, the optimal reverberation time included both zero and nonzero values. The results generally support previous theoretical results.     

Comparative intelligibility investigation of single-channel noise-reduction algorithms for Chinese, Japanese, and English

A large number of single-channel noise-reduction algorithms have been proposed based largely on mathematical principles. Most of these algorithms, however, have been evaluated with English speech. Given the different perceptual cues used by native listeners of different languages including tonal languages, it is of interest to examine whether there are any language effects when the same noise-reduction algorithm is used to process noisy speech in different languages. A comparative evaluation and investigation is taken in this study of various single-channel noise-reduction algorithms applied to noisy speech taken from three languages: Chinese, Japanese, and English. Clean speech signals (Chinese words and Japanese words) were first corrupted by three types of noise at two signal-to-noise ratios and then processed by five single-channel noise-reduction algorithms. The processed signals were finally presented to normal-hearing listeners for recognition. Intelligibility evaluation showed that the majority of noise-reduction algorithms did not improve speech intelligibility. Consistent with a previous study with the English language, the Wiener filtering algorithm produced small, but statistically significant, improvements in intelligibility for car and white noise conditions. Significant differences between the performances of noise-reduction algorithms across the three languages were observed.     

Adapting to suprasegmental lexical stress errors in foreign-accented speech

Can native listeners rapidly adapt to suprasegmental mispronunciations in foreign-accented speech? To address this question, an exposure-test paradigm was used to test whether Dutch listeners can improve their understanding of non-canonical lexical stress in Hungarian-accented Dutch. During exposure, one group of listeners heard a Dutch story with only initially stressed words, whereas another group also heard 28 words with canonical second-syllable stress (e.g., EEKhorn, "squirrel" was replaced by koNIJN "rabbit"; capitals indicate stress). The 28 words, however, were non-canonically marked by the Hungarian speaker with high pitch and amplitude on the initial syllable, both of which are stress cues in Dutch. After exposure, listeners' eye movements were tracked to Dutch target-competitor pairs with segmental overlap but different stress patterns, while they listened to new words from the same Hungarian speaker (e.g., HERsens, herSTEL, "brain," "recovery"). Listeners who had previously heard non-canonically produced words distinguished target-competitor pairs better than listeners who had only been exposed to Hungarian accent with canonical forms of lexical stress. Even a short exposure thus allows listeners to tune into speaker-specific realizations of words' suprasegmental make-up, and use this information for word recognition.