A NEW MASK-BASED OBJECTIVE MEASURE FOR PREDICTING THE INTELLIGIBILITY OF BINARY MASKED SPEECH.

Mask-based objective speech-intelligibility measures have been successfully proposed for evaluating the performance of binary masking algorithms. These objective measures were computed directly by comparing the estimated binary mask against the ground truth ideal binary mask (IdBM). Most of these objective measures, however, assign equal weight to all time-frequency (T-F) units. In this study, we propose to improve the existing mask-based objective measures by weighting each T-F unit according to its target or masker loudness. The proposed objective measure shows significantly better performance than two other existing mask-based objective measures.

Minimum spectral contrast needed for vowel identification by normal hearing and cochlear implant listeners.

The minimum spectral contrast needed for vowel identification by normal-hearing and cochlear implant listeners was determined in this study. In experiment 1, a spectral modification algorithm was used that manipulated the channel amplitudes extracted from a 6-channel continuous interleaved sampling (CIS) processor to have a 1-10 dB spectral contrast. The spectrally modified amplitudes of eight natural vowels were presented to six Med-EI/CIS-link users for identification. Results showed that subjects required a 4-6 dB contrast to identify vowels with relatively high accuracy. A 4-6 dB contrast was needed independent of the individual subject's dynamic range (range 9-28 dB). Some cochlear implant (CI) users obtained significantly higher scores with vowels enhanced to 6 dB contrast compared to the original, unenhanced vowels, suggesting that spectral contrast enhancement can improve the vowel identification scores for some CI users. To determine whether the minimum spectral contrast needed for vowel identification was dependent on spectral resolution (number of channels available), vowels were processed in experiment 2 through n (n =4, 6, 8, 12) channels, and synthesized as a linear combination of n sine waves with amplitudes manipulated to have a 1-20 dB spectral contrast. For vowels processed through 4 channels, normal-hearing listeners needed a 6 dB contrast, for 6 and 8 channels a 4 dB contrast was needed, consistent with our findings with CI listeners, and for 12 channels a 1 dB contrast was sufficient to achieve high accuracy (>80%). The above-mentioned findings with normal-hearing listeners suggest that when the spectral resolution is poor, a larger spectral contrast is needed for vowel identification. Conversely, when the spectral resolution is fine, a small spectral contrast (1 dB) is sufficient. The high identification score (82%) achieved with 1 dB contrast was significantly higher than any of the scores reported in the literature using synthetic vowels, and this can be attributed to the fact that we used natural vowels which contained duration and spectral cues (e.g., formant movements) present in fluent speech. The outcomes of experiments 1 and 2, taken together, suggest that CI listeners need a larger spectral contrast (4-6 dB) than normal-hearing listeners to achieve high recognition accuracy, not because of the limited dynamic range, but because of the limited spectral resolution.

A neural network model for optimizing vowel recognition by cochlear implant listeners.

Due to the variability in performance among cochlear implant (CI) patients, it is becoming increasingly important to find ways to optimally fit patients with speech processing strategies. This paper proposes an approach based on neural networks, which can be used to automatically optimize the performance of CI patients. The neural network model is implemented in two stages. In the first stage, a neural network is trained to mimic the CI patient's performance on the vowel identification task. The trained neural network is then used in the second stage to adjust a free parameter to improve vowel recognition performance for each individual patient. The parameter examined in this study was a weighting function applied to the compressed channel amplitudes extracted from a 6-channel continuous interleaved sampling (CIS) strategy. Two types of weighting functions were examined, one which assumed channel interaction, and one which assumed no interaction between channels. Results showed that the neural network models closely matched the performance of five Med-EI/CIS-Link implant patients. The resulting weighting functions obtained after neural network training improved vowel performance, with the larger improvement (4%) attained by the weighting function which modeled channel interaction.

Speech recognition by normal-hearing and cochlear implant listeners as a function of intensity resolution.

The importance of intensity resolution in terms of the number of intensity steps needed for speech recognition was assessed for normal-hearing and cochlear implant listeners. In experiment 1, the channel amplitudes extracted from a six-channel continuous interleaved sampling (CIS) processor were quantized into 2, 4, 8, 16, or 32 steps. Consonant recognition was assessed for five cochlear implant listeners, using the Med-El/CIS-link device, as a function of the number of steps in the electrical dynamic range. Results showed that eight steps within the dynamic range are sufficient for reaching asymptotic performance in consonant recognition. These results suggest that amplitude resolution is not a major factor in determining consonant identification. In experiment 2, the relationship between spectral resolution (number of channels) and intensity resolution (number of steps) in normal-hearing listeners was investigated. Speech was filtered through 4-20 frequency bands, synthesized as a linear combination of sine waves with amplitudes extracted from the envelopes of the bandpassed waveforms, and then quantized into 2-32 levels to produce stimuli with varying degrees of intensity resolution. Results showed that the number of steps needed to achieve asymptotic performance was a function of the number of channels and the speech material used. For vowels, asymptotic performance was obtained with four steps, while for consonants, eight steps were needed for most channel conditions, consistent with our findings in experiment 1. For sentences processed though 4 channels, 16 steps were needed to reach asymptotic performance, while for sentences processed through 16 channels, 4 steps were needed. The results with normal-hearing listeners on sentence recognition point to an inverse relationship between spectral resolution and intensity resolution. When spectral resolution is poor (i.e., a small number of channels is available) a relatively fine intensity resolution is needed to achieve high levels of understanding. Conversely, when the intensity resolution is poor, a high degree of spectral resolution is needed to achieve asymptotic performance. The results of this study, taken together with previous findings on the effect of reduced dynamic range, suggest that the performance of cochlear implant subjects is primarily limited by the small number (four to six) of channels received, and not by the small number of intensity steps or reduced dynamic range.

The effect of parametric variations of cochlear implant processors on speech understanding.

This study investigated the effect of five speech processing parameters, currently employed in cochlear implant processors, on speech understanding. Experiment 1 examined speech recognition as a function of stimulation rate in six Med-E1/CIS-Link cochlear implant listeners. Results showed that higher stimulation rates (2100 pulses/s) produced a significantly higher performance on word and consonant recognition than lower stimulation rates (<800 pulses/s). The effect of stimulation rate on consonant recognition was highly dependent on the vowel context. The largest benefit was noted for consonants in the /uCu/ and /iCi/ contexts, while the smallest benefit was noted for consonants in the /aCa/ context. This finding suggests that the /aCa/ consonant test, which is widely used today, is not sensitive enough to parametric variations of implant processors. Experiment 2 examined vowel and consonant recognition as a function of pulse width for low-rate (400 and 800 pps) implementations of the CIS strategy. For the 400-pps condition, wider pulse widths (208 micros/phase) produced significantly higher performance on consonant recognition than shorter pulse widths (40 micros/phase). Experiments 3-5 examined vowel and consonant recognition as a function of the filter overlap in the analysis filters, shape of the amplitude mapping function, and signal bandwidth. Results showed that the amount of filter overlap (ranging from -20 to -60 dB/oct) and the signal bandwidth (ranging from 6.7 to 9.9 kHz) had no effect on phoneme recognition. The shape of the amplitude mapping functions (ranging from strongly compressive to weakly compressive) had only a minor effect on performance, with the lowest performance obtained for nearly linear mapping functions. Of the five speech processing parameters examined in this study, the pulse rate and the pulse width had the largest (positive) effect on speech recognition. For a fixed pulse width, higher rates (2100 pps) of stimulation provided a significantly better performance on word recognition than lower rates (<800 pps) of stimulation. High performance was also achieved by jointly varying the pulse rate and pulse width. The above results indicate that audiologists can optimize the implant listener's performance either by increasing the pulse rate or by jointly varying the pulse rate and pulse width.

The effect of reduced dynamic range on speech understanding: implications for patients with cochlear implants.

OBJECTIVE: To determine the effect of reduced dynamic range on speech understanding when the speech signals are processed in a manner similar to a 6-channel cochlear implant speech processor. DESIGN: Signals were processed in a manner similar to a 6-channel cochlear implant processor and output as a sum of sine waves with frequencies equal to the center frequencies of the analysis filters. The amplitudes of the sine waves were compressed in a systematic fashion to simulate the effect of reduced dynamic range. The compressed signals were presented to 10 normal-hearing listeners for identification. RESULTS: There was a significant effect of compression for all test materials. The effect of the compression on speech understanding was different for the three test materials (vowels, consonants, and sentences). Vowel recognition was affected the most by the compression, and consonant recognition was affected the least by the compression. Feature analysis indicated that the reception of place information was affected the most. Sentence recognition was moderately affected by the compression. CONCLUSIONS: Dynamic range should affect the speech perception abilities of cochlear implant users. Our results suggest that a relatively wide dynamic range is needed for a high level of vowel recognition and a relatively small dynamic range is sufficient to maintain consonant recognition. We infer from this outcome that, if other factors were held equal, an implant patient with a small dynamic range could achieve moderately high scores on tests of consonant recognition but poor performance on vowel recognition, and that it is more likely for an implant patient with a large dynamic range to obtain high scores on vowel recognition than for an implant patient with a small dynamic range.

Recognition of sentences in noise by normal-hearing listeners using simulations of speak-type cochlear implant signal processors.

To assess whether more channels are needed to understand speech in noise than in quiet, we processed speech in a manner similar to that of spectral peak-like cochlear implant processors and presented it at a +2-dB signal-to-noise ratio to normal-hearing listeners for identification. The number of analysis filters varied from 8 to 16, and the number of maximum channel amplitudes selected in each cycle varied from 2 to 16. The results show that more channels are needed to understand speech in noise than in quiet, and that high levels of speech understanding can be achieved with 12 channels. Selecting more than 12 channel amplitudes out of 16 channels did not yield significant improvements in recognition performance.

Word recognition by children listening to speech processed into a small number of channels: data from normal-hearing children and children with cochlear implants.

OBJECTIVE: The aims of this study were 1) to determine the number of channels of stimulation needed by normal-hearing adults and children to achieve a high level of word recognition and 2) to compare the performance of normal-hearing children and adults listening to speech processed into 6 to 20 channels of stimulation with the performance of children who use the Nucleus 22 cochlear implant. DESIGN: In Experiment 1, the words from the Multisyllabic Lexical Neighborhood Test (MLNT) were processed into 6 to 20 channels and output as the sum of sine waves at the center frequency of the analysis bands. The signals were presented to normal-hearing adults and children for identification. In Experiment 2, the wideband recordings of the MLNT words were presented to early-implanted and late-implanted children who used the Nucleus 22 cochlear implant. RESULTS: Experiment 1: Normal-hearing children needed more channels of stimulation than adults to recognize words. Ten channels allowed 99% correct word recognition for adults; 12 channels allowed 92% correct word recognition for children. Experiment 2: The average level of intelligibility for both early- and late-implanted children was equivalent to that found for normal-hearing adults listening to four to six channels of stimulation. The best intelligibility for implanted children was equivalent to that found for normal-hearing adults listening to six channels of stimulation. The distribution of scores for early- and late-implanted children differed. Nineteen percent of the late-implanted children achieved scores below that allowed by a 6-channel processor. None of the early-implanted children fell into this category. CONCLUSIONS: The average implanted child must deal with a signal that is significantly degraded. This is likely to prolong the period of language acquisition. The period could be significantly shortened if implants were able to deliver at least eight functional channels of stimulation. Twelve functional channels of stimulation would provide signals near the intelligibility of wideband signals in quiet.

An event-related potential study of older children with an early history of failure to thrive.

Elementary and junior high school children (n = 13), who were diagnosed with nonorganic failure to thrive (FTT) as infants and toddlers, were compared with a normal control group (n = 14) on visual event-related potentials (ERPs) elicited during a primed lexical decision task. Positive stimuli were real words that were identical to the priming stimuli; negative stimuli were nonpronounceable letter strings. Although the groups did not differ in word-list reading level, the former FTT group had slower reaction (decision) times and did not show ERP evidence of priming in the N400 epoch. Anterior sites yielded better separation of the real words and letter strings than posterior sites. A late anterior component between 500 msec to 650 msec poststimulus onset showed the largest condition effect for both groups. The control group had a larger negative going late anterior component to words than the FTT group. The combined reaction time and ERP findings point to less automatized word recognition in the FTT group.

An electrophysiological study of school-aged children with a history of failure to thrive during infancy.

Sixty-five subjects, ages 8 to 12, participated in a visual electrophysiological study. Twenty-two of the subjects had received a diagnosis of nonorganic failure-to-thrive (FTT) before the age of three. The remaining 43 subjects had no history of FTT and served as Controls. IQs were obtained with the abbreviated WISC-III, and the Controls were split into two groups, LO IQ and HI IQ, to provide a LO IQ Control group with an average IQ equivalent to the FTT group. Event-related brain potentials (ERPs) were recorded from five scalp locations during a cued continuous performance task (CPT). Subjects had to press a button every time they saw the letter "X" following the letter "A" (50 targets out of 400 stimuli). During the CPT, the FTT subjects made marginally more errors of omission to targets than the LO IQ Control group and significantly more errors of omission than the HI IQ Control subjects. The groups did not differ significantly on errors of commission (false alarms) or reaction times to targets. ERP averages revealed a group difference in amplitude in a late slow wave for the 50 non-X stimuli (false targets) that followed the letter A. This difference was greatest over frontal sites, where the FTT group had a more negative going slow wave than each control group. Late frontal negativity to No Go stimuli has been linked with post-decisional processing, notably in young children. Thus, the FTT subjects may have less efficient inhibitory processes, reflected by additional late frontal activation.

On the number of channels needed to understand speech.

Recent studies have shown that high levels of speech understanding could be achieved when the speech spectrum was divided into four channels and then reconstructed as a sum of four noise bands or sine waves with frequencies equal to the center frequencies of the channels. In these studies speech understanding was assessed using sentences produced by a single male talker. The aim of experiment 1 was to assess the number of channels necessary for a high level of speech understanding when sentences were produced by multiple talkers. In experiment 1, sentences produced by 135 different talkers were processed through n (2 < or = n < or = 16) number of channels, synthesized as a sum of n sine waves with frequencies equal to the center frequencies of the filters, and presented to normal-hearing listeners for identification. A minimum of five channels was needed to achieve a high level (90%) of speech understanding. Asymptotic performance was achieved with eight channels, at least for the speech material used in this study. The outcome of experiment 1 demonstrated that the number of channels needed to reach asymptotic performance varies as a function of the recognition task and/or need for listeners to attend to fine phonetic detail. In experiment 2, sentences were processed through 6 and 16 channels and quantized into a small number of steps. The purpose of this experiment was to investigate whether listeners use across-channel differences in amplitude to code frequency information, particularly when speech is processed through a small number of channels. For sentences processed through six channels there was a significant reduction in speech understanding when the spectral amplitudes were quantized into a small number (< 8) of steps. High levels (92%) of speech understanding were maintained for sentences processed through 16 channels and quantized into only 2 steps. The findings of experiment 2 suggest an inverse relationship between the importance of spectral amplitude resolution (number of steps) and spectral resolution (number of channels).

Human fetuses have nonlinear cardiac dynamics.

Approximate entropy (ApEn) is a statistic that quantifies regularity in time series data, and this parameter has several features that make it attractive for analyzing physiological systems. In this study, ApEn was used to detect nonlinearities in the heart rate (HR) patterns of 12 low-risk human fetuses between 38 and 40 wk of gestation. The fetal cardiac electrical signal was sampled at a rate of 1,024 Hz by using Ag-AgCl electrodes positioned across the mother's abdomen, and fetal R waves were extracted by using adaptive signal processing techniques. To test for nonlinearity, ApEn for the original HR time series was compared with ApEn for three dynamic models: temporally uncorrelated noise, linearly correlated noise, and linearly correlated noise with nonlinear distortion. Each model had the same mean and SD in HR as the original time series, and one model also preserved the Fourier power spectrum. We estimated that noise accounted for 17.2-44.5% of the total between-fetus variance in ApEn. Nevertheless, ApEn for the original time series data still differed significantly from ApEn for the three dynamic models for both group comparisons and individual fetuses. We concluded that the HR time series, in low-risk human fetuses, could not be modeled as temporally uncorrelated noise, linearly correlated noise, or static filtering of linearly correlated noise.

High vagal tone is associated with more efficient regulation of homeostasis in low-risk human fetuses.

Homeostasis is maintained primarily by the parasympathetic nervous system and is thought to provide a physiological substrate for the development of complex behaviors. This investigation was undertaken to test the hypothesis that infants with high parasympathetic tone are more efficient regulators of homeostasis than infants with low parasympathetic tone. Respiratory sinus arrhythmia (RSA) was used as a measure of parasympathetic tone, and the efficiency of homeostatic control was quantified, for each infant, by the slope (SRSA) and correlation coefficient (RRSA) of the regression line relating fluctuations in heart period and fluctuations in RSA. To test our hypothesis, we examined the relationship between RSA and both SRSA and RRSA in 34 low-risk human fetuses between 36 and 40 weeks gestation. We found that fetuses who were parasympathetic-dominated had larger SRSA and RRSA values, and hence were more efficient regulators of homeostasis, than fetuses who were sympathetic-dominated. The results of our analyses are important because they establish, very early in development, a physiological basis for the relationship between vagal tone and the development of complex behaviors.

Introduction to cochlear implants.

Cochlear implants have been very successful in restoring partial hearing to profoundly deaf people. Many individuals with implants are now able to communicate and understand speech without lip-reading, and some are able to talk over the phone. Children with implants can develop spoken-language skills and attend normal schools (i.e., schools with normal-hearing children). The greatest benefits with cochlear implantation have occurred in patients who (1) acquired speech and language before their hearing loss, and (2) have shorter duration of deafness. Gradual, but steady, improvements in speech production and speech perception have also occurred in prelingually deafened adults or children.

The recognition of sentences in noise by normal-hearing listeners using simulations of cochlear-implant signal processors with 6-20 channels.

Sentences were processed through simulations of cochlear-implant signal processors with 6, 8, 12, 16, and 20 channels and were presented to normal-hearing listeners at +2 db S/N and at -2 db S/N. The signal-processing operations included bandpass filtering, rectification, and smoothing of the signal in each band, estimation of the rms energy of the signal in each band (computed every 4 ms), and generation of sinusoids with frequencies equal to the center frequencies of the bands and amplitudes equal to the rms levels in each band. The sinusoids were summed and presented to listeners for identification. At issue was the number of channels necessary to reach maximum performance on tests of sentence understanding. At +2 dB S/N, the performance maximum was reached with 12 channels of stimulation. At -2 dB S/N, the performance maximum was reached with 20 channels of stimulation. These results, in combination with the outcome that in quiet, asymptotic performance is reached with five channels of stimulation, demonstrate that more channels are needed in noise than in quiet to reach a high level of sentence understanding and that, as the S/N becomes poorer, more channels are needed to achieve a given level of performance.

The identification of speech in noise by cochlear implant patients and normal-hearing listeners using 6-channel signal processors.

OBJECTIVE: To compare the recognition of vowels and sentences in noise by cochlear implant patients using a 6-channel, continuous interleaved sampling (CIS) processor and by normal-hearing subjects listening to speech processed in the manner of the implant processor and output as six amplitude-modulated sine waves. DESIGN: Subjects, 11 normal-hearing listeners and 7 cochlear implant patients, were presented natural vowels produced by men, women, and girls in /hVd/ context and sentences from the Hearing In Noise Test (HINT) lists at +15, +10, and +5 dB signal to noise ratio (SNR) for identification. Stimuli for the normal-hearing subjects were preprocessed through a simulation of a 6-channel implant processor and were output as the sum of sinusoids at the center frequencies of the analysis filters. RESULTS: For the multitalker vowels, four of the seven patients achieved scores within +/-1 standard deviation of the mean for normal-hearing listeners at +15 and +10 dB SNR. At the +5 dB SNR three patients achieved scores within +/-1 standard deviation of the mean for the normal-hearing listeners. For the HINT sentences, four of seven patients achieved scores within +/-1 standard deviation of the mean for the normal-hearing listeners at +15 dB and at +10 dB SNR and two achieved scores within that range at +5 dB SNR. CONCLUSION: Our results extend the range of stimulus conditions, from quiet to modest amounts of noise, in which the CIS strategy allows the best performing patients to extract most, if not all, of the information available to normal-hearing subjects listening to speech processed into six channels.

The identification of consonants and vowels by cochlear implant patients using a 6-channel continuous interleaved sampling processor and by normal-hearing subjects using simulations of processors with two to nine channels.

OBJECTIVE: To compare the vowel and consonant identification ability of cochlear implant patients using a 6-channel continuous interleaved sampling (CIS) processor and of normal-hearing subjects using simulations of processors with two to nine channels. DESIGN: Subjects, 10 normal-hearing listeners and seven cochlear implant patients, were presented synthetic vowels in /bVt/ context, natural vowels produced by men, women, and girls in /hVd/ context, and consonants in /aCa/ context for identification. Stimuli for the normal-hearing subjects were pre-processed through simulations of implant processors with two to nine channels and were output as the sum of sinusoids at the center frequencies of the analysis filters. RESULTS: Five implant patients' scores fell within the range of normal performance with a 6-channel processor when the patients were tested with synthetic vowels. Four patients' scores fell within the range of normal with a 6-channel processor when the patients were tested with multitalker vowels. Five patients' scores fell within the range of normal for a 6-channel processor for the consonant feature "place of articulation." CONCLUSION: Signal processing technology for cochlear implants has matured sufficiently to allow some patients who use CIS processors and a small number of monopolar electrodes to achieve scores on tests of speech identification that are within the range of scores established by normal-hearing subjects listening to speech processed through a small number of channels.

The recognition of vowels produced by men, women, boys, and girls by cochlear implant patients using a six-channel CIS processor.

Five patients who used a six-channel, continuous interleaved sampling (CIS) cochlear implant were presented vowels, in two experiments, from a large sample of men, women, boys, and girls for identification. At issue in the first experiment was whether vowels from one speaker group, i.e., men, were more identifiable than vowels from other speaker groups. At issue in the second experiment was the role of the fifth and sixth channels in the identification of vowels from the different speaker groups. It was found in experiment 1 that (i) the vowels produced by men were easier to identify than vowels produced by any of the other speaker groups, (ii) vowels from women and boys were more difficult to identify than vowels from men but less difficult than vowels from girls, and (iii) vowels from girls were more difficult to identify than vowels from all other groups. In experiment 2 removal of channels 5 and 6 from the processor impaired the identification of vowels produced by women, boys and girls but did not impair the identification of vowels produced by men. The results of experiment 1 demonstrate that scores on tests of vowels produced by men overestimate the ability of patients to recognize vowels in the broader context of multi-talker communication. The results of experiment 2 demonstrate that channels 5 and 6 become more important for vowel recognition as the second formants of the speakers increase in frequency.