Brian R. Shelton

Speech bandwidth extension method and apparatus

A speech bandwidth extension method and apparatus analyzes narrowband speech sampled at 8 kHz using LPC analysis to determine its spectral shape and inverse filtering to extract its excitation signal. The excitation signal is interpolated to a sampling rate of 16 kHz and analyzed for pitch control and power level. A white noise generated wideband signal is then filtered to provide a synthesized wideband excitation signal. The narrowband shape is determined and compared to templates in respective vector quantizer codebooks, to select respective highband shape and gain. The synthesized wideband excitation signal is then filtered to provide a highband signal which is, in turn, added to the narrowband signal, interpolated to the 16 kHz sample rate, to produce an artificial wideband signal. The apparatus may be implemented on a digital signal processor chip.

Comparison of three adaptive psychophysical procedures

Auditory thresholds were experimentally determined using three adaptive procedures. The procedures are a simple staircase procedure, a maximum‐likelihood technique and PEST. The threshold of a sinusoid in noise (simultaneous task) and the threshold of a sinusoid in a forward masking experiment (successive task) were investigated. Number of trials per block, for the staircase and maximum‐likelihood procedures, and stopping rule, for PEST, were varied in four different conditions. Nearly equal numbers of trials were obtained for each procedure. The observed variability in successive determinations of threshold was used to estimate standard errors of measurement. All three procedures produced standard errors of measurement that were nearly equal. Some slight differences among the procedures may be of interest for particular applications.

Psychophysical tuning curves independent of signal level.

Psychophysical tuning curves were obtained for three subjects in a forward masking paradigm, with signal levels of 21, 30, and 50 dB SPL at 1000 Hz and 30, 40, and 50 dB SPL with a 3000-Hz signal. All tuning curves were measured in quiet and with a background noise adjusted to produce a constant signal-to-noise ratio of E/N0 = 16 dB. The results indicated that the tuning curves obtained in quite change shape with increasing signal level, whereas the shapes of tuning curves measured in noise are the same at all signal levels. A control condition demonstrates that only signal energy in a narrow band near the signal frequency is useful in detecting the signal. This control along with the invariance of tuning curve shape with signal level support the interpretation that the broadening of tuning curves with increased signal level, observed when the signals are presented in quiet, is caused by the presence of off-frequency components.

A reversal of lateralization, with images appearing on the side of the delay

Unlike the case with tones, where sensitivity to interaural delay exists only for low frequencies, delays in the envelopes of amplitude modulated sounds are effective, regardless of the frequency of the carrier. Interestingly, for pulse‐modulation of band‐limited noise, the percept caused by interaural delay can be quite misleading, with the image appearing on the side of the lagging signal [Hafter and Ricard, J. Acoust. Soc. Am. 53, 335(A) (1973)]. This anomaly depends upon the frequency of the carder, with the image being on the “correct” side for low frequencies and then exhibiting a damped oscillation around the midline as the carrier is increased. A simple model which fits the data well associates the location of the image with the center of gravity (CG) of the cross‐correlation function, computed from the acoustic waveforms with the signs of the correlation maintained in the computation. In this way, negative peaks in the function act to repel the CG just as positive peaks act to attract it. For loc...

The lateralization of carrier‐delayed and modulation‐delayed pulse‐modulated signals

An anomalous result has been reported in the lateralization of pulse‐modulated signals presented with a modulation delay: For a fixed interaural modulation delay the spatial position of the binaural image varies with the carrier frequency, and for some carrier frequencies the image is heard to be closer to the ear presented with the delayed modulation waveform. A cross‐correlation model has been proposed to account for these results [Hafter, Shelton, and Green, J. Acoust, Soc. Am. 68, S16 (1980)]. The lateralization of pulse‐modulated signals presented with an interaural carrier delay is reported here. The data do not contain “illusory” results, that is, the binaural image is always heard closer to the ear presented with a carrier advance. However, variations in performance as a function of carrier frequency correspond to the dependency of modulation‐delay performance on carrier frequency. The assumption that a cross‐correlation process is involved in the binaural analysis of both modulation delay and car...

Presentation‐order effects in the discrimination of rapidly presented short intervals

We previously reported a forward masking effect in a duration discrimination task [G. J. Fleet and B. R. Shelton, J. Acoust. Soc. Am. Suppl. 1 74, S10 (1983)], suggesting that the restriction of performance in the discrimination of rapidly presented brief durations was peripheral. This would predict a presentation‐order effect, where a long‐short discrimination would be easier than short‐long. Two experiments examined this idea. The first study was two‐interval forced‐choice paradigm, with the thresholds for the long‐short and short‐long presentations adapted separately. A second experiment examined these same discriminations in a same‐different paradigm. Signals were 80‐dB SPL noise bursts fast gated in a continuous 60‐dB noise background. Base durations of 25, 50, 100, and 200 ms were used, with ISI values of 25, 50, 100, 200, 400, 800, and 1600 ms. Counter to expectations, thresholds were much lower for the short‐long presentations. These data are discussed in terms of their relation to the proposed pe...

A comparison of the relative merits of four psychophysical procedures

The efficiency of four psychophysical procedures to estimate the detection threshold of a brief 1000‐Hz sinusoid presented in noise was assessed in naive observers. Six threshold estimates were obtained for ten subjects in 50‐trial adaptive runs using each of four procedures: two‐alternative adaptive staircase, two‐alternative maximum‐likelihood, three‐alternative adaptive staircase and the three‐alternative maximum‐likelihood method. The four procedures produced equivalent thresholds, although the staircase procedures provided biased measurements in the initial runs. A trial‐by‐trial analysis of the variability of threshold estimates revealed a marked advantage to the use of three alternatives with the adaptive staircase, but not with the maximum‐likelihood procedures. The data provide information regarding the relative efficiency of the four procedures, and recommend that particular procedures should be used in specific instances. In general, the commonly used two‐alternative adaptive staircase is the l...

Forward masking tuning curves obtained in the presence of a backward noise masker

The combined effect of a backward and forward mask has often been noted to be difficult to predict on the basis of the masking obtained from either masker alone. This report describes the nonlinear effects of a backward mask on the shape of forward masking tuning curves. Five‐point psychophysical tuning curves were obtained in two observers, using 200‐ms forward maskers of 500, 750, 1000, 1250, and 1500 Hz, and a 10‐ms, 1000‐Hz signal presented at 40 dB SPL. The forward mask had a 2‐ms rise/fall time and the signal was fast‐gated 3 ms after the forward mask. The backward mask was a 100‐ms burst of white noise, presented 3 ms after the signal offset, with a fast rise/fall. Tuning curves were obtained in quiet, and with backward maskers of 20, 50,and 80 dB SPL. Tuning curves became sharper as the level of the backward mask was increased. The tip of the tuning curve dropped by at least 20 dB in the presence of the 80‐dB backward mask, and by less than 10 dB at other frequencies. The results are discussed in terms of the strategies employed by listeners to detect a signal in the presence of a forward mask. [Work supported by NSERC.]

A demonstration of peripheral constraint on an auditory duration discrimination task

An earlier study investigated the discrimination of auditory durations presented at a rapid rate in a two‐alternative forced‐choice task. The results indicated that a constant Weber fraction was obtained across base durations from 25–1600 ms if a long interstimulus interval (ISI) separated the two alternative presentations, but performance degraded with short base durations if a short ISI was used [B. R. Shelton, J. Acoust. Soc. Am. Suppl. 1 72, S89 (1982)]. Here, the same measurements were taken under two presentation conditions: (1) a binaural presentation of both intervals, and (2) a monaural presentation of the two intervals, each to a different ear. Base durations of 25, 50, 100, and 200 ms were used, with ISI values of 25, 50, 100, 200, 400, 800, and 1600 ms. The binaurally presented condition replicated the previous results, but the presentation of intervals to alternate ears produced a marked improvement in performance, especially with short ISI presentations. This finding suggests that the discri...

Duration discrimination of noise bursts presented at a rapid rate

The influence of the presentation rate of two noise bursts on the discrimination of duration was assessed using an adaptive two‐alternative, forced‐choice procedure. The time difference required for reliable discrimination was determined for 70 dB SPL noise bursts of 25‐, 50‐, 100‐, 200‐, 400‐, 800‐, and 1600‐ms duration. The interval between the presentation of the two noise bursts in each trial was set at either 50, 200, or 1600 ms. The Weber fraction, Δt/t, was determined by an interaction between the noise duration and the rate of presentation. With a presentation interval of 1600 ms, Δt/t was a constant of 0.08 at all noise durations. Presentation intervals of 50 and 200 ms produced Δt/t functions which were elevated for short noise durations, but converged on the 0.08 Δt/t performance level with noise duration of 1600 ms. The Δt/t for a 25‐ms signal was 0.30 with a 50‐ms presentation interval, and 0.22 with 200 ms between noise bursts. The results are discussed in relation to neural constraints on t...