An individual differences model for multidimensional scaling is outlined in which individuals are assumed differentially to weight the several dimensions of a common “psychological space”. A corresponding method of analyzing similarities data is proposed, involving a generalization of “Eckart-Young analysis” to decomposition of three-way (or higher-way) tables. In the present case this decomposition is applied to a derived three-way table of scalar products between stimuli for individuals. This analysis yields a stimulus by dimensions coordinate matrix and a subjects by dimensions matrix of weights. This method is illustrated with data on auditory stimuli and on perception of nations.

Analysis of individual differences in multidimensional scaling via an n-way generalization of 'eckart-young' decomposition

Pregnant women recited a particular speech passage aloud each day during their last 6 weeks of pregnancy. Their newborns were tested with an operant-choice procedure to determine whether the sounds of the recited passage were more reinforcing than the sounds of a novel passage. The previously recited passage was more reinforcing. The reinforcing value of the two passages did not differ for a matched group of control subjects. Thus, third-trimester fetuses experience their mothers' speech sounds and that prenatal auditory experience can influence postnatal auditory preferences.

Prenatal maternal speech influences newborns' perception of speech sounds

The effects of perceptual learning of talker identity on the recognition of spoken words and sentences were investigated in three experiments. In each experiment, listeners were trained to learn a set of 10 talkers’ voices and were then given an intelligibility test to assess the influence of learning the voices on the processing of the linguistic content of speech. In the first experiment, listeners learned voices from isolated words and were then tested with novel isolated words mixed in noise. The results showed that listeners who were given words produced by familiar talkers at test showed better identification performance than did listeners who were given words produced by unfamiliar talkers. In the second experiment, listeners learned novel voices from sentence-length utterances and were then presented with isolated words. The results showed that learning a talker’s voice from sentences did not generalize well to identification of novel isolated words. In the third experiment, listeners learned voices from sentence-length utterances and were then given sentence-length utterances produced by familiar and unfamiliar talkers at test. We found that perceptual learning of novel voices from sentence-length utterances improved speech intelligibility for words in sentences. Generalization and transfer from voice learning to linguistic processing was found to be sensitive to the talker-specific information available during learning and test. These findings demonstrate that increased sensitivity to talker-specific information affects the perception of the linguistic properties of speech in isolated words and sentences.

/pdf/talker-specific-learning-in-speech-perception-5f4o498x32.pdf

Talker-specific learning in speech perception

Successful speech perception requires that listeners map the acoustic signal to linguistic categories. These mappings are not only probabilistic, but change depending on the situation. For example, one talker's /p/ might be physically indistinguishable from another talker's /b/ (cf. lack of invariance). We characterize the computational problem posed by such a subjectively nonstationary world and propose that the speech perception system overcomes this challenge by (a) recognizing previously encountered situations, (b) generalizing to other situations based on previous similar experience, and (c) adapting to novel situations. We formalize this proposal in the ideal adapter framework: (a) to (c) can be understood as inference under uncertainty about the appropriate generative model for the current talker, thereby facilitating robust speech perception despite the lack of invariance. We focus on 2 critical aspects of the ideal adapter. First, in situations that clearly deviate from previous experience, listeners need to adapt. We develop a distributional (belief-updating) learning model of incremental adaptation. The model provides a good fit against known and novel phonetic adaptation data, including perceptual recalibration and selective adaptation. Second, robust speech recognition requires that listeners learn to represent the structured component of cross-situation variability in the speech signal. We discuss how these 2 aspects of the ideal adapter provide a unifying explanation for adaptation, talker-specificity, and generalization across talkers and groups of talkers (e.g., accents and dialects). The ideal adapter provides a guiding framework for future investigations into speech perception and adaptation, and more broadly language comprehension.

Robust speech perception: recognize the familiar, generalize to the similar, and adapt to the novel.

This paper presents a survey of automatic speaker recognition techniques. The paper indudes a discussion of the speaker-dependent properties of the speech signal, methods for selecting an efficient set of speech measurements, results of experimental studies illustrating the performance of various methods of speaker recognition, and a comparision of the performance of automatic methods with that of human listeners. Both text-dependent as well as text-independent speaker-recognition techniques are discussed.

Automatic recognition of speakers from their voices

Sixteen listeners attempted to identify the talker when listening to speech samples of varying duration and content. The samples, recorded by 10 different talkers, were of five types: excerpted vowels, excerpted consonant‐vowel (CV) sequences, monosyllabic words, disyllabic nonsense words, and sentences. Identification accuracy improved directly with the number of phonemes in the sample even when duration was controlled. Stimulus‐response matrices differed substantially between the two vowels ([a] and [i]) used in the vowel and CV samples: relative identifiability of the talkers, response preference, and error patterns were all found to depend on vowel type. Confusion matrices for a given vowel exhibit definite asymmetries. In a limited additional study, subsets of listeners made identifying responses with the tapes reversed; performance deteriorated on even the briefest excerpts. The results pose some difficulties for a model of talker‐identification behavior based on attributes of voice quality.

Effects of stimulus content and duration on talker identification.

chapter 9 – Speaker Recognition

This Letter describes a technique for making digital measurements of the average speech level during speech spurts. Results are free of variability owing to the subjectivity and unreliability of the human reader. Data relating these measurements to VU‐meter readings and to loudness judgments are presented.

Technique for Objective Measurement of Speech Levels

Twenty subjects used a sorting apparatus to form hierarchically related clusters, based on similarity, of 24 three‐parameter frequency‐modulated tones (the HC task; duration: 15 min). Another partially overlapping group made judgments of similarity of all 552 ordered pairs of the same tones (the PC task; duration: 60 min). Both sets of data were analyzed by a new multidimensional scaling method (due to J. D. Carroll and J‐J. V. Chang) that accommodates individual differences in weights assigned to the unique dimensions of a group space. The structure based on PC data recovered the generating parameters (modulation frequency, deviation extent, and modulation waveform) and revealed some interaction but little scaling or intersubject difference. Stimuli were less uniformly distributed in the HC structure, and subjects differed considerably, both in the scaling of modulation frequency and in the assignment of weights to dimensions. HC and PC data from the same individual are not equivalent; each provides its own type of information about perceived similarity. However, averaging HC data over subjects and scaling by the Shepard‐Kruskal technique produces a structure that can be rotated into close correspondence with the PC structure.

Comparison of Sorting and Pairwise Similarity Judgment Techniques for Scaling Auditory Stimuli

Sixteen listeners identified the voices represented by various samples of the utterances of 10 talkers. Types of samples and their percentages correct were sentences (98%), disyllables (87%), monosyllables (81%), excerpted C‐V sequences (62%), and excerpted vowels (55%). Another group of naive listeners performed ABX discriminations among all pairs of six of these talkers using sentences with the X sentence in reverse. Some of the more significant findings were (1) when identification was based on a single excerpted vowel, both the order of identifiability and the pattern of confusions were affected by the particular vowel chosen; (2) reinstating the consonant before the vowel raised performance in general and altered the response pattern for some talkers; (3) confusion matrices were asymmetrical for both naive and sophisticated listeners. In general, talker identification depends strongly on what was said. These findings offer little encouragement for the notion that the recognition process is based on v...

Peter D. Bricker

Papers

Effects of stimulus content and duration on talker identification.

chapter 9 – Speaker Recognition

Technique for Objective Measurement of Speech Levels

Comparison of Sorting and Pairwise Similarity Judgment Techniques for Scaling Auditory Stimuli

Temporal Effects in Talker Identification