Models Of Pitch Perception Research Articles

Comparison of microstructures in absolute threshold and in the pitch-intensity relation—a search for a place theory of pitch perception

We have made fine-grained measurements of absolute threshold and of the pitch-intensity relation for sine tones as functions of frequency. For some subjects the measured microstructures have small enough error and are sufficiently stable in time that one can hope to draw conclusions from the data. For some ears, in some frequency ranges, there is little threshold microstructure. In these cases the pitch-intensity relation tends to show a frequency-independent shift. More commonly, however, there is considerable threshold microstructure, as much as 10 dB over one semitone. In those cases there is inevitably appreciable microstructure in the pitch-intensity relation; the pitch shift may change sign six or seven times per octave. We have attempted to exploit the microstructures to test models of pitch perception for sine tones. We made standard assumptions about neural driven firing rate. We also supposed that absolute threshold indicates neural sensitivity. We found that our microstructure data are inconsistent with models in which pitch is derived from either the peak locus or the centroid of an average neural firing rate pattern. This conclusion is in agreement with a previous study of pitch shifts caused by a leading tone. [W. M. Hartmann, J. Acoust. Soc. Am. Suppl. 1 68, S109 (1980)]. [Work supported by the NSF and by the NIH.]

The spectral shape of the dominance region for ripple noise pitch

The repetition pitch and pitch strength of ripple noise depends on information in a spectral region located three to five times the reported pitch. A quantitative measure of the shape of this spectral dominance region for repetition pitch was obtained using a pitch strength, discrimination procedure. Listeners discriminated between two pitches of ripple noise. The pitch strength of the two stimuli was decreased until the listeners were at threshold in their ability to make the pitch discrimination. The pitch strength required for threshold was determined as a function of low‐ or high‐pass filtering the ripple noise using filters with a variety of cutoff frequencies. This function described the spectral dominance region for a variety of ripple noise conditions. The dominance region had a bandpass characteristic with a center frequency at four times the reported pitch, and the low‐frequency roll off in pitch strength was shallower than the high‐frequency roll off. The shape of the dominance region was proportional to the repetition pitch of the ripple noise. Thus, pitch strength was strongest when the spectrum contained energy at four times the reported pitch, weaker when only high‐frequency energy was present and weakest when only low‐frequency energy was present. These results will be discussed in terms of the recent models of pitch perception, especially the Peripheral Weighting Model. [Work supported by NSF.]

Is pitch a learned attribute of sounds? Two points in support of Terhardt’s pitch theory

Terhardt [J. Acoust. Soc. Am. 55, 1061-1069 (1974)] postulated a pitch perception model wherein a learning stage constitutes an integral part: it is only repeated exposure to patterns of spectral pitch that will generate the percept of virtual pitch (i.e., the residue). Two examples, one clinical and one musical, are cited to support the idea that perception of the pitch of complex tones represents a case of pattern perception which is acquired with experience.

On the shift in pitch of a sine tone caused by a preceding tone

The pitch of a brief target sine tone is changed by a preceding tone. The pitch of the target tone is shifted away from the pitch of the preceding tone [A. Jaroszewski and A. Rakowski, Acustica 34, 220 (1976); W. M. Hartmann and B. J. Blumenstock. J. Acoust. Soc. Am. 60, S40 (A) (1976)]. Recent experiments on this effect show the following: (1) The shift is present for tones near 1000 Hz and near 200 Hz and for tones heard at 80 dB and at 40 dB. (2) Smaller shifts exist when 40 dB preceding and target tones are in opposite ears, and for preceding tones 40 dB less intense than the target tones. (3) The shift tends to vanish for missing-fundamental preceding tones with no spectral energy near the target tone. These experiments support a model of pitch perception based upon a central neural excitation pattern which is an intensity-compressed representation of the stimulus spectrum. The results can be understood from a model of short-term neural adaptation and a model for the extraction of pitch from a neural excitation pattern. [Supported by the NSF.]

Discrimination of pitch in minimal duration tonal stimuli

Discrimination of pitch for very brief tonal stimuli was measured in a two interval forced choice experiment in which experienced subjects were required to select the stimulus complex that contained acoustic events more disparate in pitch. Within each complex a comparatively long (750 ms) standard tone was followed by a 750 ms interstimulus interval and a brief comparison tone of variable frequency and duration (2, 4, 6. or 8 cycles). Frequencies used for the standard tone were 125, 250, 500, 1000, and 2000 Hz. All stimuli were matched for loudness. Analysis of performance indicates an interaction between number of cycles in the brief tone sample and frequency of the standard tone. Patterns of errors are considered in relation to spectral and periodicity models of pitch perception.

Pitch strength of filtered ripple noise: evidence for a dominant pitch region?

Ripple noise is produced when a broadband noise is delayed (T) and either added to or subtracted from the original noise. The resulting waveform has a power spectrum in which power varies cosinusoidally as a function of frequency. Ripple noise has been used to study pitch perception and echo processing. In the present experiments subjects were asked to discriminate between two noises which differed in the amount of delay introduced (one stimulus contained a delay, T, the second stimulus a delay, T + 0.1T). The discriminability of these two ripple noises was studied as a function of the amount of attenuation added to the delayed noise. Increases in attenuation of the delayed noise decreases the strength of the pitch of ripple noise. Delays (T) of 1, 2, 5, and 10 msec were used to produce ripple noises which were passed through very steep, one-third octave, digital filters centered at 500, 1000, 2000, and 4000 Hz. The pitch of the ripple noise produced with any delay (T) was strongest (i.e., the greatest amount of attenuation was added to the delayed noise before discrimination performance was near threshold) when the center frequency of the filters were at four times 1/T. For center frequencies lower or higher than 4×1/T the pitch was weaker. The pitch of the ripple noise when the delayed noise was substracted from the undelayed noise was weaker than when the delayed noise was added to the undelayed noise. These data will be discussed in terms of recent models of pitch perception and the assumption that pitch is determined in a dominant spectral region. [Work supported by NSF.]

Octave convergence in melodic perception

Previous research indicates that the pitch of a single tone can be conceptualized as a bidimensional quantity, reflecting the overall pitch level of the tone (tone height) and its position in the octave (tone chroma). It has been suggested that the dimension of tone chroma is irrelevant to perception of the component tones of a melody. Supporting this argument, perception of a melody is disrupted when its tones are displaced by octave multiples. This has been interpreted as showing that melodic perception is based upon the magnitude of successive pitch intervals. However, octave displacement violates both interval magnitude and direction. In the current study, melodies were subjected to structural transformations in which the tones were displaced by octave multiples, either preserving or violating the direction of successive pitch intervals (melodic contour). Replicating previous research, when melodic contour was violated performance was severely disrupted. When contour was preserved, however, the melodies were identified as accurately when the tones were displaced by octave multiples as when the melodies were untransformed. These results demonstrate that the bidimensional model of pitch perception applies to melodies, and that contour plays a significant role in melodic perception. [Work supported by NIMH.]

Pronounced binaural pitch phenomenon

Using a special multiple-phase-shift filter with a flat amplitude response, a rather strong sensation of pitch was evoked by dichotic presentation of white Gaussian noise. The existence of this pitch, which resembles a periodicity pitch, can be understood on the basis of the ’’central spectrum’’ model of pitch perception [Bilsen and Goldstein, J. Acoust. Soc. Am. 55, 292–296 (1974)]. Subject Classification: [43]65.54, [43]65.62.

Evaluation of pitch-strength predictions of the pattern-transformation model of pitch

The pattern-transformation model of pitch perception [Wightman, J. Acoust. Soc. Am. 54, 407–416 (1973)] can successfully predict the pitch of many types of complex-tone stimuli. In addition, the model provides numerical estimates of pitch “strength.” In order to test these latter predictions we assumed that performance in a musical interval-recognition paradigm [Houtsma and Goldstein, J. Acoust. Soc. Am. 51, 520–529 (1972)] is a direct indicator of pitch strength; the better the performance, the stronger the pitch. Thus, Houtsma and Goldstein's data can be used to evaluate the pitch-strength predictions of the model To do this, a set of stimuli was defined such that all stimuli in the set led to the same level of performance in the interval-recognition paradigm; the stimuli lay on one of Houtsma and Goldstein's “constant-performance contours.” Then pitch-strength predictions were generated for these stimuli. If our assumption is valid, all the stimuli should have the same predicted pitch strength. Except for stimuli of low fundamental frequencies, the predictions follow this pattern. The failure with low-fundamental stimuli seems to be a result of our initial assumption that spectral resolution (critical bandwidth) is a constant percentage of frequency (i. e., that the peripheral analyzer has constant Q). If the constant Q assumption is relaxed, so that our assumed critical bandwidths agree more closely with available data, the performance of the model is much improved.

The effects of relative phase and the number of components on residue pitch

The pitch of the residue produced by 6- and 12-component waves whose components were either in cosine or random phase was measured as a function of the frequency region of the components. The components were all of equal amplitude and the frequency spacing between them was 200 Hz. Waves with the same lowest component were found to produce the same pitch: neither the number of components nor the relative phase of the components was important in determining the pitch of the residue. If every component in a wave was a multiple of 200 Hz, the wave produced a 200 Hz pitch; if such a set of components was shifted in frequency by a small amount, there was a corresponding linear shift in the pitch of the residue. When the lowest component in the wave was below about 900 Hz, the slope values associated with the lines relating the pitch shift to the frequency shift varied little from their average value of 0.21. As the lowest component increased from 900 to 2580 Hz the slope values decreased to about 0.08. These findings are in good agreement with the current model of pitch perception based on interpeak durations.

Prediction of Cubic Difference Tones from a Model of Pitch Perception

Using a model of pitch perception already shown to give accurate predictions of residue pitches and difference tones [83rd Meeting, J. Acoust. Soc. Amer. 52, 141(A) 0972)], we have been able to predict cubic difference tones for frequency ratios of the primary tones from 1.167 to 1.500. The model involves the determination of axis crossing intervals of the stimulus at an appropriate basilar membrane (BM) location through Flanagan's BM transfer function, taking into account both amplitude and phase changes. The model notes and marks the time of significant differences in adjacent axis crossing intervals of the signal at the appropriate BM location. Time intervals corresponding to cubic difference tones are found between these time marks. In almost all cases investigated, the axis crossing intervals must be determined at a BM location corresponding to the lower primary frequency. Contrary to our earlier statement, the prediction of cubic difference tones from this model is achieved without any assumption of nonlinear distortion mechanisms. [Work supported by NIH.]

Discrimination of fundamental frequency contours in synthetic speech: implications for models of pitch perception.

The just-noticeable difference (JND) for selected aspects of voice fundamental frequency (F0) contours was determined by varying the F0 control parameter of a digitally simulated terminal analog speech synthesizer. Data were obtained from three subjects for a number of 250-msec segments of the synthetic vowel /ɛ/ differing only in fundamental frequency. Results indicate that the subjects can detect a change of 0.3 Hz in a constant F0 contour when F0 = 120 Hz, but the JND is an order of magnitude larger (2.0 Hz) when the F0 contour is a linear descending ramp (32 Hz/sec). Sensitivity to rate of change of F0 in linear ramps is surprisingly good; greatest sensitivity occurs when one ramp increases and the other decreases (JND = 12 Hz/sec). High-pass filtering of the stimuli improves performance slightly, suggesting that the fundamental component is not involved in the detection of changes in F0. Substitution of the synthetic stimulus /ya/ with its dynamic formant contours in place of /ɛ/ degrades performance only very slightly. Implications of these data for models of pitch perception mechanisms are discussed.

A Model for Residue Pitch

We have developed a model of pitch perception that accurately predicts all subjective (residue) pitches measured in the experiments of Schouten, Ritsma, and Cardozo [J. Acoust. Soc. Amer. 34, 1418–1424 (1962)]. The primary assumptions of the model are (1) that the transfer function relating amplitude of basilar membrane displacement at a particular membrane location to acoustic pressure external to the ear represents a bandpass filter of low selectivity and frequency-dependent phase change; (2) that the pattern of axis crossing intervals of the signal representing basilar membrane displacement at any membrane location is neurologically encoded and analyzed; (3) that there is a mechanism capable of detecting a variation of approximately 0.3% in the duration of axis crossing intervals for the frequency range considered. Prediction of residue pitches requires no assumption of nonlinear distortion anywhere in the filtering process, nor does it require the presence of any amplitude encoding mechanism such as an envelope detector. The model also predicts perception of the difference tone without any appeal to nonlinear processes. Prediction of more complex combination tones does require the assumption of nonlinear distortion in the filtering process. The model gives at least qualitatively satisfactory explanations for other auditory phenomena such as the critical band and masking. [Work supported by NIH.]