Modeling the Development of Phonetic Representations

语音表示的发展建模

• 批准号: ES/R006660/1
• 负责人: Sharon Goldwater
• 金额: $ 37.61万
• 依托单位: University of Edinburgh
• 依托单位国家: 英国
• 项目类别: Research Grant
• 财政年份: 2018
• 资助国家: 英国
• 起止时间: 2018 至 无数据
• 项目状态: 已结题
• 来源: https://gtr.ukri.org/projects?ref=ES%2FR006660%2F1
• 关键词: Modeling; Development; Phonetic; Representations

Listeners differ cross-linguistically in the cues, or perceptual dimensions, they rely on when perceiving speech. For example, Japanese listeners categorizing English [l] and [r] do not rely on the same acoustic features of the speech signal (e.g., the third formant) that native English listeners do. These cross-linguistic differences are typically attributed to listeners' knowledge of sound categories. For example, English listeners know that [l] and [r] are two categories, whereas Japanese listeners know that [l] and [r] are part of the same category; and this knowledge is hypothesized to affect their reliance on dimensions.The proposed research tests the hypothesis that category knowledge is not necessary for perceptual dimension learning to occur. Drawing on representation learning methods that have performed well in low-resource automatic speech recognition, where extensive labeled training data are not available, two models are proposed that learn dimensions without relying on knowledge of sound categories. The first relies on temporal information as a proxy for category knowledge, while the second relies on top-down information from similar words, which infants have been shown to use. These models are evaluated on their ability to predict listeners' discrimination judgments from speech perception experiments on native and non-native contrasts, when trained on the same language background as the listeners.

听众在感知语音时所依赖的线索或感知维度在跨语言方面存在差异。例如，对英语 [l] 和 [r] 进行分类的日本听众并不依赖于与英语母语听众相同的语音信号声学特征（例如，第三共振峰）。这些跨语言差异通常归因于听众对声音类别的了解。例如，英语听众知道 [l] 和 [r] 是两个类别，而日语听众知道 [l] 和 [r] 属于同一类别；假设这些知识会影响他们对维度的依赖。所提出的研究测试了这样的假设：类别知识对于感知维度学习来说并不是必需的。利用在资源匮乏的自动语音识别中表现良好的表示学习方法（在这种情况下，无法获得大量标记的训练数据），提出了两种模型，可以在不依赖声音类别知识的情况下学习维度。第一个依赖于时间信息作为类别知识的代理，而第二个依赖于婴儿已被证明使用的相似单词的自上而下的信息。当在与听众相同的语言背景下进行训练时，这些模型根据对母语和非母语对比的语音感知实验来预测听众的辨别判断的能力进行评估。

项目成果

Infant Phonetic Learning as Perceptual Space Learning: A Crosslinguistic Evaluation of Computational Models

• DOI: 10.1111/cogs.13314
• 发表时间: 2023-07
• 期刊: Cognitive science
• 影响因子: 2.5
• 作者: Yevgen Matusevych;Thomas Schatz;H. Kamper;Naomi H Feldman;S. Goldwater

Input matters in the modeling of early phonetic learning

输入在早期语音学习建模中很重要

• 发表时间: 2020
• 期刊: Proceedings of the Annual Conference of the Cognitive Science Society
• 作者: Li, Ruolan;Schatz, Thomas;Matusevych, Yevgen;Goldwater, Sharon;Feldman, Naomi H.
• 通讯作者: Feldman, Naomi H.

Multilingual Acoustic Word Embedding Models for Processing Zero-resource Languages

• DOI: 10.1109/icassp40776.2020.9054202
• 发表时间: 2020-02
• 期刊: ICASSP 2020 - 2020 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)
• 作者: H. Kamper;Yevgen Matusevych;S. Goldwater

Improved Acoustic Word Embeddings for Zero-Resource Languages Using Multilingual Transfer

• DOI: 10.1109/taslp.2021.3060805
• 发表时间: 2020-06
• 期刊: IEEE/ACM Transactions on Audio, Speech, and Language Processing
• 作者: H. Kamper;Yevgen Matusevych;S. Goldwater

Do Infants Really Learn Phonetic Categories?

• DOI: 10.1162/opmi_a_00046
• 发表时间: 2021
• 期刊: Open mind : discoveries in cognitive science
• 作者: Feldman NH;Goldwater S;Dupoux E;Schatz T
• 通讯作者: Schatz T