Alpha activity marking word boundaries mediates speech segmentation

2012 ◽  
Vol 36 (12) ◽  
pp. 3740-3748 ◽  
Author(s):  
Antoine J. Shahin ◽  
Mark A. Pitt
Keyword(s):  
Alpha Activity ◽  
Speech Segmentation ◽  
Word Boundaries

Bootstrapping Word Boundaries: A Bottom-up Corpus-Based Approach to Speech Segmentation

1997 ◽  
Vol 33 (2) ◽  
pp. 111-153 ◽  
Author(s):  
Paul Cairns ◽  
Richard Shillcock ◽  
Nick Chater ◽  
Joe Levy
Keyword(s):  
Speech Segmentation ◽  
Bottom Up ◽  
Word Boundaries

scholarly journals When statistics collide: The use of transitional and phonotactic probability cues to word boundaries by Brazilian-Portuguese adults

2019 ◽  
Author(s):  
Rodrigo Dal Ben ◽  
Débora de Hollanda Souza ◽  
Jessica Hay
Keyword(s):  
Second Language Acquisition ◽  
Brazilian Portuguese ◽  
Speech Segmentation ◽  
Transitional Probability ◽  
Phonotactic Probability ◽  
Test Items ◽  
Early Language Development ◽  
Significant Difference ◽  
Word Boundaries ◽  
Statistical Regularities

Statistical regularities in linguistic input shape early language development and second language acquisition. For example, both transitional probability and phonotactic probability play a role in speech segmentation, however, it remains unclear whether or how these statistics are combined when small differences in phonotactic probabilities are presented. We conducted two experiments to investigate the effects of transitional and phonotactic probabilities on speech segmentation by Brazilian-Portuguese-speaking adults. Four pseudo-languages, with six words each, were created. The transitional probabilities between words’ biphones were high, whereas the probabilities between part-words’ biphones were lower. Although the within and between word phonotactic probability were always high, they varied slightly across the familiarization languages and test words/part-words. Languages 1 and 2 had familiarization words with unbalanced phonotactics, but target words and part-words used at test were phonotactically balanced. Languages 3 and 4 had familiarization words with balanced phonotactics, but phonotactics were unbalanced across test items; In Language 3 words had slightly lower phonotactics that part-words. The reverse was true for Language 4. Eighty-one Brazilian-Portuguese speaking adults were divided in four groups. Each group was familiarized with one version of the language and then tested on two-alternative forced choice trials. Participants presented with Languages 1, 2 and 4 preferred words to part-words at test. However, participants who heard Language 3 did not select words above chance. There was no significant difference in word selection between Language 4 and Languages 1 and 2, despite the fact that phonotactics were higher during both familiarization and test for words from the fourth language. These findings indicate that phonotactic and transitional information can be tracked and combined to facilitate or impair speech segmentation. Furthermore, they suggest that subtle differences in phonotactics are more informative of word boundaries than congruency between high phonotactic and transitional probability cues.

scholarly journals Active Listening

2020 ◽  
Author(s):  
Karl J. Friston ◽  
Noor Sajid ◽  
David Ricardo Quiroga-Martinez ◽  
Thomas Parr ◽  
Cathy J. Price ◽  
...  
Keyword(s):  
Active Vision ◽  
Acoustic Signals ◽  
Generative Models ◽  
Speech Segmentation ◽  
Face Validity ◽  
Active Listening ◽  
Unified Framework ◽  
Prior Beliefs ◽  
Word Boundaries ◽  
Selection Of

AbstractThis paper introduces active listening, as a unified framework for synthesising and recognising speech. The notion of active listening inherits from active inference, which considers perception and action under one universal imperative: to maximise the evidence for our (generative) models of the world. First, we describe a generative model of spoken words that simulates (i) how discrete lexical, prosodic, and speaker attributes give rise to continuous acoustic signals; and conversely (ii) how continuous acoustic signals are recognised as words. The ‘active’ aspect involves (covertly) segmenting spoken sentences and borrows ideas from active vision. It casts speech segmentation as the selection of internal actions, corresponding to the placement of word boundaries. Practically, word boundaries are selected that maximise the evidence for an internal model of how individual words are generated. We establish face validity by simulating speech recognition and showing how the inferred content of a sentence depends on prior beliefs and background noise. Finally, we consider predictive validity by associating neuronal or physiological responses, such as the mismatch negativity and P300, with belief updating under active listening, which is greatest in the absence of accurate prior beliefs about what will be heard next.

Harmonic cues for speech segmentation: a cross-linguistic corpus study on child-directed speech

2013 ◽  
Vol 41 (2) ◽  
pp. 439-461 ◽  
Author(s):  
F. NIHAN KETREZ
Keyword(s):  
Word Segmentation ◽  
Vowel Harmony ◽  
Speech Segmentation ◽  
Natural Languages ◽  
Corpus Study ◽  
Artificial Languages ◽  
Word Boundaries

ABSTRACTPrevious studies on the role of vowel harmony in word segmentation are based on artificial languages where harmonic cues reliably signal word boundaries. In this corpus study run on the data available at CHILDES, we investigated whether natural languages provide a learner with reliable segmentation cues similar to the ones created artificially. We observed that in harmonic languages (child-directed speech to thirty-five Turkish and three Hungarian children), but not in non-harmonic ones (child-directed speech to one Farsi and four Polish children), harmonic vowel sequences are more likely to appear within words, and non-harmonic ones mostly appear across word boundaries, suggesting that natural harmonic languages provide a learner with regular cues that could potentially be used for word segmentation along with other cues.

THE EXPLOITATION OF SUBPHONEMIC ACOUSTIC DETAIL IN L2 SPEECH SEGMENTATION

2014 ◽  
Vol 36 (4) ◽  
pp. 709-731 ◽  
Author(s):  
Ellenor Shoemaker
Keyword(s):  
Native Speakers ◽  
First Language ◽  
Speech Segmentation ◽  
Phonological Acquisition ◽  
First Language Transfer ◽  
Allophonic Variation ◽  
L2 Phonology ◽  
Word Boundaries ◽  
French Speaking

The current study addresses an aspect of second language (L2) phonological acquisition that has received little attention to date—namely, the acquisition of allophonic variation as a word boundary cue. The role of subphonemic variation in the segmentation of speech by native speakers has been indisputably demonstrated; however, the acquisition of allophonic cues in L2 phonology remains underexplored. We examine here L2 learners’ acquisition and perception of noncontrastive acoustic differentiation at word boundaries in English. Fifty French-speaking students of English were tested on their ability to differentiate potentially ambiguous phrases in which word boundaries are marked by the word-initial aspiration of plosives (e.g.,Lou stopsvs.loose tops) or prevocalic glottal stops (e.g.,tea matvs.team at). Participants showed greater sensitivity to the presence of glottal stops than aspiration, suggesting that glottal stops may represent a more perceptually salient segmentation cue for learners than aspiration. We discuss the implications of these results regarding the role of first language transfer versus the universality of some segmentation cues.

scholarly journals The use of tonal coarticulation in segmentation of artificial language speech: A study with Mandarin listeners

2021 ◽  
pp. 1-25
Author(s):  
Zhe-Chen Guo ◽  
Shu-Chen Ou
Keyword(s):  
Speech Segmentation ◽  
Artificial Language ◽  
Null Effect ◽  
Word Boundaries ◽  
Statistical Regularities ◽  
Prosodic Boundaries ◽  
Cue Redundancy

Abstract Tonal carryover assimilation, whereby a tone is assimilated to the preceding one, is conditioned by prosodic boundaries in a way suggesting that its presence may signal continuity or lack of a boundary. Its possibility as a speech segmentation cue was investigated in two artificial language (AL) learning experiments. Mandarin-speaking listeners identified the “words” of a three-tone AL (e.g., [pé.tī.kù]) after listening to six long speech streams in which the words were repeated continuously without pauses. The first experiment revealed that segmentation was disrupted in an “incongruent-cues” condition where tonal carryover assimilation occurred across AL word boundaries and conflicted with statistical regularities in the speech streams. Segmentation was neither facilitated nor inhibited in a “congruent-cues” condition where tonal carryover assimilation occurred only within the AL words in 27% of the repetitions and never across word boundaries. A null effect was again found for the congruent-cues condition of the second experiment, where all AL word repetitions carried tonal carryover assimilation. These findings show that tonal carryover assimilation is exploited to resolve segmentation problems when cues conflict. Its null effect in the congruent-cues conditions might be linked to cue redundancy and suggest that it is weighted low in the segmentation cue hierarchy.

Spectral and Topographic Microstructure of Brain Alpha Activity During Drowsiness at Sleep Onset and REM Sleep

2000 ◽  
Vol 14 (3) ◽  
pp. 151-158 ◽  
Author(s):  
José Luis Cantero ◽  
Mercedes Atienza
Keyword(s):  
Rem Sleep ◽  
Spectral Component ◽  
Sleep Onset ◽  
Quantitative Information ◽  
Maximum Energy ◽  
Alpha Activity ◽  
The Other ◽  
Classification Algorithms ◽  
Microstructural Properties ◽  
Brain States

Abstract High-resolution frequency methods were used to describe the spectral and topographic microstructure of human spontaneous alpha activity in the drowsiness (DR) period at sleep onset and during REM sleep. Electroencephalographic (EEG), electrooculographic (EOG), and electromyographic (EMG) measurements were obtained during sleep in 10 healthy volunteer subjects. Spectral microstructure of alpha activity during DR showed a significant maximum power with respect to REM-alpha bursts for the components in the 9.7-10.9 Hz range, whereas REM-alpha bursts reached their maximum statistical differentiation from the sleep onset alpha activity at the components between 7.8 and 8.6 Hz. Furthermore, the maximum energy over occipital regions appeared in a different spectral component in each brain activation state, namely, 10.1 Hz in drowsiness and 8.6 Hz in REM sleep. These results provide quantitative information for differentiating the drowsiness alpha activity and REM-alpha by studying their microstructural properties. On the other hand, these data suggest that the spectral microstructure of alpha activity during sleep onset and REM sleep could be a useful index to implement in automatic classification algorithms in order to improve the differentiation between the two brain states.

Sign in / Sign up

Export Citation Format

Share Document