Learning to perceive speech: how fricative perception changes,and how it stays the same

A part of becoming a mature perceiver involves learning what signal properties provide relevant information about objects and events in the environment. Regarding speech perception, evidence supports the position that allocation of attention to various signal properties changes as children gain experience with their native language, and so learn what information is relevant to recognizing phonetic structure in that language. However, one weakness in that work has been that data have largely come from experiments that all use similarly designed stimuli and show similar age-related differences in labeling. In this study, two perception experiments were conducted that used stimuli designed differently from past experiments, with different predictions. In experiment 1, adults and children (4, 6, and 8 years of age) labeled stimuli with natural /f/ and /[see text]/ noises and synthetic vocalic portions that had initial formant transitions varying in appropriateness for /f/ or /[see text]/. The prediction was that similar labeling patterns would be found for all listeners. In experiment 2, adults and children labeled stimuli with initial /s/-like and /[see text]/-like noises and synthetic vocalic portions that had initial formant transitions varying in appropriateness for /s/ or /[see text]/. The prediction was that, as found before, children would weight formant transitions more and fricative noises less than adults, but that this age-related difference would elicit different patterns of labeling from those found previously. Results largely matched predictions, and so further evidence was garnered for the position that children learn which properties of the speech signal provide relevant information about phonetic structure in their native language.     

Superconformal invariants or how to relate four-point AdS amplitudes

Using the form of N=2$N=2$ superconformal invariants we derive the explicit relation between the bottom and top components of the correlator of four stress-tensor multiplets in N=4$N=4$ super-Yang–Mills. The result is given in terms of an eighth order differential operator acting on the function of two variables which characterises these correlators. It allows us to show a non-trivial consistency relation between the known results for the corresponding supergravity amplitudes on AdS₅${\mathit{AdS}}_5$.     

Fine-grained pitch processing of music and speech in congenital amusia

Congenital amusia is a lifelong disorder of music processing that has been ascribed to impaired pitch perception and memory. The present study tested a large group of amusics (n=17) and provided evidence that their pitch deficit affects pitch processing in speech to a lesser extent: Fine-grained pitch discrimination was better in spoken syllables than in acoustically matched tones. Unlike amusics, control participants performed fine-grained pitch discrimination better for musical material than for verbal material. These findings suggest that pitch extraction can be influenced by the nature of the material (music vs speech), and that amusics' pitch deficit is not restricted to musical material, but extends to segmented speech events.     

Comment on "Measurement of pitch in speech: an implementation of Goldstein's theory of pitch perception"

The pitch detection algorithm proposed by Duifhuis, Willems, and Sluyter [J. Acoust. Soc. Am. 71, 1568-1580 (1982)] can be made more than 20 times faster by replacing the harmonic sieve procedure by the approximate common denominator procedure, the results differing only slightly.     

The speed of pitch resolution in a musical context.

In five experiments, we investigated the speed of pitch resolution in a musical context. In experiments 1-3, listeners were presented an incomplete scale (doh, re, mi, fa, sol, la, ti) and then a probe tone. Listeners were instructed to make a rapid key-press response to probe tones that were relatively proximal in pitch to the last note of the scale (valid trials), and to ignore other probe tones (invalid trials). Reaction times were slower if the pitch of the probe tone was dissonant with the expected pitch (i.e., the completion of the scale, or doh) or if the probe tone was nondiatonic to the key implied by the scale. In experiments 4 and 5, listeners were presented a two-octave incomplete arpeggio, and then a probe tone. In this case, listeners were asked to make a rapid key-press response to probe tones that were relatively distant in pitch from the last note of the arpeggio. Under these conditions, registral direction and pitch proximity were the dominant influences on reaction time. Results are discussed in view of research on auditory attention and models of musical pitch.     

A connectionist study on the role of pitch in infant-directed speech

Infant-directed speech (IDS) is believed to facilitate language learning. However, the benefit may be either due to clearer acoustic correlates to linguistic structures, or simply increased attention from infants induced by IDS exaggerated prosody. This study investigated the pure effect of IDS pitch on lexical tone learning, with attentional/affective factors removed by using artificial neural networks. Following training with the pitch of Mandarin tones in IDS versus adult-directed speech, the networks yielded equal tonal categorization for both registers. IDS pitch produced no additional linguistic support. IDS pitch appears to strictly play the non-linguistic role of attention/affect, which may indirectly benefit learning.     

基于基音参数规整及统计分布模型距离的语音情感识别

提出一种根据基音提取的频率分辨率确定自适应窗口的改进Parzen窗方法估计基音概率密度,兼顾了基音统计分布模型在低频段的高分辨率和高频段的平滑;提出利用不同性别的基音分布规律的性别区分算法,对于长句可以达到98%的识别率;通过分析基音均值、方差、统计分布模型在性别上的差异,对基音参数进行基于性别差异的规整;引入规整后的基音均值和基音方差,以及基音统计分布模型距离作为情感特征参数;最后利用K最近邻方法对汉语情感语料进行识别。利用常规方法提取的参数最后得到的识别率为73.8%,而使用经过性别差异规整的基音参数和基音统计分布距离的识别率提高到81%。     

汉语口语与朗读话语陈述句音高比较

汉语语句通常存在音高下倾现象,然而关于语句内部韵律词的具体音高表现目前的研究尚较欠缺。本研究使用的对话语料选自973电话语料库,包括69段对话,涉及79位说话人;朗读话语语料为广播电台两位主持人的新闻播音,长度为221个语句,对语句内部韵律词的高音点、低音点及音域进行了分析,结果显示对话与朗读话语多数语句的音高呈前高后低的走势,不过口语对话较长语句前半段的音高下降趋势不太明显。与朗读话语相比,口语对话韵律词的音域通常比较小。对话语句最后一个韵律词的音域相对较大,而朗读话语内部韵律词的音域大多没有差异。本研究的结果,将有助于语音合成中语句内部韵律词音阶及音域的构拟。     

A comparison on the pitch of statement of dialogue and read speech in Chinese Putonghua

Experimental evidence showed that declination exists in most sentences in Chinese Putonghua.However,the specific pitch variation of prosodic words(PWs) within sentences has not been fully understood.The dialogue material used in this study was taken from the 973 telephone conversation corpus,which includes 69 dialogues,with 79 speakers involved.The read speech material was recordings of news announcement by two announcers from radio, totally 221 sentences.Top and bottom points of pitch contour and the range of prosodic words with the sentences were studied.It was found that,for both dialogue and read speech,pitch declination exists for most sentences,with minor exceptions in the first part of longer sentences for dialogue.Compared to read speech,the pitch range of prosodic words for dialogue is smaller. For dialogue,the pitch range of the last prosodic words of the sentences are relatively larger, while there is no significant difference for pitch ranges of prosodic words in most sentences of read speech.This study will be helpful for the modeling of the pitch range and register of prosodic words in sentences in speech synthesis.     

Second generation wavelet transform-based pitch period estimation and voiced/unvoiced decision for speech signals

Pitch detection is an important part of speech recognition and speech processing. In this paper, a pitch detection algorithm based on second generation wavelet transform was developed. The proposed algorithm reduces the computational load of those algorithms that were based on classical wavelet transform. The proposed pitch detection algorithm was tested for both real speech and synthetic speech signal. Some experiments were carried out under noisy environment condition to evaluate the accuracy and robustness of the proposed algorithm. Results showed that the proposed algorithm was robust to noise and provided accurate estimates of the pitch period for both low-pitched and high-pitched speakers. Moreover, different wavelet filters that were obtained using second generation wavelet transform were considered to see the effects of them on the proposed algorithm. It was noticed that Haar filter showed good performance as compared to the other wavelet filters.     

En route to Background Independence: Broken split-symmetry,and how to restore it with bi-metric average actions

The most momentous requirement a quantum theory of gravity must satisfy is Background Independence, necessitating in particular an ab initio derivation of the arena all non-gravitational physics takes place in, namely spacetime. Using the background field technique, this requirement translates into the condition of an unbroken split-symmetry connecting the (quantized) metric fluctuations to the (classical) background metric. If the regularization scheme used violates split-symmetry during the quantization process it is mandatory to restore it in the end at the level of observable physics. In this paper we present a detailed investigation of split-symmetry breaking and restoration within the Effective Average Action (EAA) approach to Quantum Einstein Gravity (QEG) with a special emphasis on the Asymptotic Safety conjecture. In particular we demonstrate for the first time in a non-trivial setting that the two key requirements of Background Independence and Asymptotic Safety can be satisfied simultaneously. Carefully disentangling fluctuation and background fields, we employ a ‘bi-metric’ ansatz for the EAA and project the flow generated by its functional renormalization group equation on a truncated theory space spanned by two separate Einstein–Hilbert actions for the dynamical and the background metric, respectively. A new powerful method is used to derive the corresponding renormalization group (RG) equations for the Newton- and cosmological constant, both in the dynamical and the background sector. We classify and analyze their solutions in detail, determine their fixed point structure, and identify an attractor mechanism which turns out instrumental in the split-symmetry restoration. We show that there exists a subset of RG trajectories which are both asymptotically safe and split-symmetry restoring: In the ultraviolet they emanate from a non-Gaussian fixed point, and in the infrared they loose all symmetry violating contributions inflicted on them by the non-invariant functional RG equation. As an application, we compute the scale dependent spectral dimension which governs the fractal properties of the effective QEG spacetimes at the bi-metric level. Earlier tests of the Asymptotic Safety conjecture almost exclusively employed ‘single-metric truncations’ which are blind towards the difference between quantum and background fields. We explore in detail under which conditions they can be reliable, and we discuss how the single-metric based picture of Asymptotic Safety needs to be revised in the light of the new results. We shall conclude that the next generation of truncations for quantitatively precise predictions (of critical exponents, for instance) is bound to be of the bi-metric type.     

应对说话人位置突变的鲁棒语声去混响*

多通道线性预测算法是一种有效的语声去混响方法,但若房间内说话人位置发生变化,房间冲激响应也会随之改变,从而导致去混响效果变差。本文研究了基于递归最小二乘算法的自适应多通道线性预测语声去混响方法,提出了一种去混响过程中检测说话人位置突变的方法,它借助已有的多通道传声器信号,并未明显增加系统开销。实验表明该方法可以有效检测说话人位置的突变,并通过重置滤波器改善算法的收敛速度和稳定性。     

Infant pitch perception: evidence for responding to pitch categories and the missing fundamental

While numerous studies on infant perception have demonstrated the infant's ability to discriminate sounds having different frequencies, little research has evaluated more sophisticated pitch perception abilities such as perceptual constancy and perception of the missing fundamental. In the present study 7-8-month-old infants demonstrated the ability to discriminate harmonic complexes from two pitch categories that differed in pitch by approximately 20% (e.g., 160 vs 200 Hz). Using a visually reinforced conditioned head-turning paradigm, a number of spectrally different tonal complexes that contained varying harmonic components but signaled the same two pitch categories were presented. After learning the basic pitch discrimination, the same infants learned to categorize spectrally different tonal complexes according to the pitches signaled by their fundamental frequencies. That is, the infants showed evidence of perceptual constancy for the pitch of harmonic complexes. Finally, infants heard tonal complexes that signaled the same pitch categories but for which the fundamental frequency was removed. Infants were still able to categorize the harmonic complexes according to their pitch categories. These results suggest that by 7 months of age infants show fairly sophisticated pitch perception abilities similar to those demonstrated by adults.     

Influence of rate of change of frequency on the overall pitch of frequency-modulated tones

The mechanism(s) determining pitch may assign less weight to portions of a sound where the frequency is changing rapidly. The present experiments explored the possible effect of this on the overall pitch of frequency-modulated sounds. Pitch matches were obtained between an adjustable unmodulated sinusoid and a sinusoidal carrier that was frequency modulated using a highly asymmetric function with the form of a repeating U or inverted U shaped function. The amplitude was constant during the 400-ms presentation time of each stimulus, except for 10-ms raised-cosine onset and offset ramps. In experiment 1, the carrier level was 50 dB SPL and the geometric mean of the instantaneous frequency of the modulated carrier, fc, was either 0.5, 1, 2, or 8 kHz. The modulation rate (fm) was 5, 10, or 20 Hz. The overall depth (maximum to minimum) of the FM was 8% of fc. For all carrier frequencies, the matched frequency was shifted away from the mean carrier frequency, downwards for the U shaped function stimuli and upwards for the repeated inverted U shaped function stimuli. The shift was typically slightly greater than 1% of fc, and did not vary markedly with fc. The effect of fm was small, but there was a trend for the shifts to decrease with increasing fm for fc = 0.5 kHz and to increase with increasing fm for fc = 2 kHz. In experiment 2, the carrier level was reduced to 20 dB SL and matches were obtained only for fc = 2 kHz. Shifts in matched frequency of about 1% were still observed, but the trend for the shifts to increase with increasing fm no longer occurred. In experiment 3, matches were obtained for a 4-kHz carrier at 50 dB SPL. Shifts of about 1% again occurred, which did not vary markedly with fm. The shifts in matched frequency observed in all three experiments are not predicted by models based on the amplitude- or intensity-weighted average of instantaneous frequency (EWAIF or IWAIF). The shifts (and the pitch shifts observed earlier for two-tone complexes and for stimuli with simultaneous AM and FM) are consistent with a model based on the assumption that the overall pitch of a frequency-modulated sound is determined from a weighted average of period estimates, with the weight attached to a given estimate being inversely related to the short-term rate of change of period and directly related to a compressive function of the amplitude.     

Enhancement of temporal cues to pitch in cochlear implants: effects on pitch ranking

The abilities to hear changes in pitch for sung vowels and understand speech using an experimental sound coding strategy (eTone) that enhanced coding of temporal fundamental frequency (F0) information were tested in six cochlear implant users, and compared with performance using their clinical (ACE) strategy. In addition, rate- and modulation rate-pitch difference limens (DLs) were measured using synthetic stimuli with F0s below 300 Hz to determine psychophysical abilities of each subject and to provide experience in attending to rate cues for the judgment of pitch. Sung-vowel pitch ranking tests for stimuli separated by three semitones presented across an F0 range of one octave (139-277 Hz) showed a significant benefit for the experimental strategy compared to ACE. Average d-prime (d') values for eTone (d' = 1.05) were approximately three time larger than for ACE (d' = 0.35). Similar scores for both strategies in the speech recognition tests showed that coding of segmental speech information by the experimental strategy was not degraded. Average F0 DLs were consistent with results from previous studies and for all subjects were less than or equal to approximately three semitones for F0s of 125 and 200?Hz.