An official website of the United States government
Direct acoustics-to-word (A2W) systems for end-to-end automatic speech recognition are simpler to train, and more efficient to decode with, than sub-word systems. However, A2W systems can have difficulties at training time when data is limited, and at decoding time when recognizing words outside the training vocabulary. To address these shortcomings, we investigate the use of recently proposed acoustic and acoustically grounded word embedding techniques in A2W systems. The idea is based on treating the final pre-softmax weight matrix of an AWE recognizer as a matrix of word embedding vectors, and using an externally trained set of word embeddings to improve the quality of this matrix. In particular we introduce two ideas: (1) Enforcing similarity at training time between the external embeddings and the recognizer weights, and (2) using the word embeddings at test time for predicting out-of-vocabulary words. Our word embedding model is acoustically grounded, that is it is learned jointly with acoustic embeddings so as to encode the words’ acoustic-phonetic content; and it is parametric, so that it can embed any arbitrary (potentially out-of-vocabulary) sequence of characters. We find that both techniques improve the performance of an A2W recognizer on conversational telephone speech.
more »
« less
Analyzing autoencoder-based acoustic word embeddings
Recent studies have introduced methods for learning acoustic word embeddings (AWEs)—fixed-size vector representations of words which encode their acoustic features. Despite the widespread use of AWEs in speech processing research, they have only been evaluated quantitatively in their ability to discriminate between whole word tokens. To better understand the applications of AWEs in various downstream tasks and in cognitive modeling, we need to analyze the representation spaces of AWEs. Here we analyze basic properties of AWE spaces learned by a sequence-to-sequence encoder-decoder model in six typologically diverse languages. We first show that these AWEs preserve some information about words’ absolute duration and speaker. At the same time, the representation space of these AWEs is organized such that the distance between words’ embeddings increases with those words’ phonetic dissimilarity. Finally, the AWEs exhibit a word onset bias, similar to patterns reported in various studies on human speech processing and lexical access. We argue this is a promising result and encourage further evaluation of AWEs as a potentially useful tool in cognitive science, which could provide a link between speech processing and lexical memory.
more »
« less
- Award ID(s):
- 1734245
- PAR ID:
- 10182737
- Date Published:
- Journal Name:
- ICLR Workshop on Bridging AI and Cognitive Science
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Segmental models are sequence prediction models in which scores of hypotheses are based on entire variable-length segments of frames. We consider segmental models for whole-word ("acoustic-to-word") speech recognition, with the feature vectors defined using vector embeddings of segments. Such models are computationally challenging as the number of paths is proportional to the vocabulary size, which can be orders of magnitude larger than when using subword units like phones. We describe an efficient approach for end-to-end whole-word segmental models, with forward-backward and Viterbi decoding performed on a GPU and a simple segment scoring function that reduces space complexity. In addition, we investigate the use of pre-training via jointly trained acoustic word embeddings (AWEs) and acoustically grounded word embeddings (AGWEs) of written word labels. We find that word error rate can be reduced by a large margin by pre-training the acoustic segment representation with AWEs, and additional (smaller) gains can be obtained by pre-training the word prediction layer with AGWEs. Our final models improve over prior A2W models.more » « less
-
Abstract Partial speech input is often understood to trigger rapid and automatic activation of successively higher-level representations of words, from sound to meaning. Here we show evidence from magnetoencephalography that this type of incremental processing is limited when words are heard in isolation as compared to continuous speech. This suggests a less unified and automatic word recognition process than is often assumed. We present evidence from isolated words that neural effects of phoneme probability, quantified by phoneme surprisal, are significantly stronger than (statistically null) effects of phoneme-by-phoneme lexical uncertainty, quantified by cohort entropy. In contrast, we find robust effects of both cohort entropy and phoneme surprisal during perception of connected speech, with a significant interaction between the contexts. This dissociation rules out models of word recognition in which phoneme surprisal and cohort entropy are common indicators of a uniform process, even though these closely related information-theoretic measures both arise from the probability distribution of wordforms consistent with the input. We propose that phoneme surprisal effects reflect automatic access of a lower level of representation of the auditory input (e.g., wordforms) while the occurrence of cohort entropy effects is task sensitive, driven by a competition process or a higher-level representation that is engaged late (or not at all) during the processing of single words.more » « less
-
null (Ed.)Previous research suggests that individuals with weaker receptive language show increased reliance on lexical information for speech perception relative to individuals with stronger receptive language, which may reflect a difference in how acoustic-phonetic and lexical cues are weighted for speech processing. Here we examined whether this relationship is the consequence of conflict between acoustic-phonetic and lexical cues in speech input, which has been found to mediate lexical reliance in sentential contexts. Two groups of participants completed standardized measures of language ability and a phonetic identification task to assess lexical recruitment (i.e., a Ganong task). In the high conflict group, the stimulus input distribution removed natural correlations between acoustic-phonetic and lexical cues, thus placing the two cues in high competition with each other; in the low conflict group, these correlations were present and thus competition was reduced as in natural speech. The results showed that 1) the Ganong effect was larger in the low compared to the high conflict condition in single-word contexts, suggesting that cue conflict dynamically influences online speech perception, 2) the Ganong effect was larger for those with weaker compared to stronger receptive language, and 3) the relationship between the Ganong effect and receptive language was not mediated by the degree to which acoustic-phonetic and lexical cues conflicted in the input. These results suggest that listeners with weaker language ability down-weight acoustic-phonetic cues and rely more heavily on lexical knowledge, even when stimulus input distributions reflect characteristics of natural speech input.more » « less
-
Abstract Speech processing often occurs amid competing inputs from other modalities, for example, listening to the radio while driving. We examined the extent to which dividing attention between auditory and visual modalities (bimodal divided attention) impacts neural processing of natural continuous speech from acoustic to linguistic levels of representation. We recorded electroencephalographic (EEG) responses when human participants performed a challenging primary visual task, imposing low or high cognitive load while listening to audiobook stories as a secondary task. The two dual-task conditions were contrasted with an auditory single-task condition in which participants attended to stories while ignoring visual stimuli. Behaviorally, the high load dual-task condition was associated with lower speech comprehension accuracy relative to the other two conditions. We fitted multivariate temporal response function encoding models to predict EEG responses from acoustic and linguistic speech features at different representation levels, including auditory spectrograms and information-theoretic models of sublexical-, word-form-, and sentence-level representations. Neural tracking of most acoustic and linguistic features remained unchanged with increasing dual-task load, despite unambiguous behavioral and neural evidence of the high load dual-task condition being more demanding. Compared to the auditory single-task condition, dual-task conditions selectively reduced neural tracking of only some acoustic and linguistic features, mainly at latencies >200 ms, while earlier latencies were surprisingly unaffected. These findings indicate that behavioral effects of bimodal divided attention on continuous speech processing occur not because of impaired early sensory representations but likely at later cognitive processing stages. Crossmodal attention-related mechanisms may not be uniform across different speech processing levels.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Conference Paper:
- The DOI is not currently available.
