Fixed-dimensional acoustic embeddings of variable-length segments in low-resource settings

By Keith D. Levin, Katharine Henry, Aren Jansen, Karen Livescu. 2013

136 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Representation & Self-Supervised Learning, Speech & Audio, Signal Processing

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Representation & Self-Supervised Learning

In Speech & Audio

In Signal Processing

Citations in the corpus

Speech2Vec: A Sequence-to-Sequence Framework for Learning Word Embeddings from Speech (2018, background): “Researchers have also explored the concept of learning vector representations from speech [9, 10, 11, 12, 13, 14].”
A segmental framework for fully-unsupervised large-vocabulary speech recognition (2016, methodology): “We also use a computationally much simpler embedding approach based on down-sampling [25].”
Audio Word2Vec: Unsupervised Learning of Audio Segment Representations Using Sequence-to-Sequence Autoencoder (2016, background): “In query-by-example spoken term detection (STD), by representing each word segment as a vector, easier indexing makes the retrieval much more efﬁcient [9, 10, 11].”
Unsupervised Cross-Modal Alignment of Speech and Text Embedding Spaces (2018, background): “The first one, referred to as Bayesian embedded segmental Gaussian mixture model (BES-GMM) [35], is a probabilistic model that represents potential word segments as fixed-dimensional acoustic word embeddings [23], and builds a whole-word acoustic model in this embedding space while jointly doing seg”
An embedded segmental K-means model for unsupervised segmentation and clustering of speech (2017, methodology): “KMeans follows as closely as possible that of BES-GMM in [22]. Both use uniform downsampling as embedding function f e: a segment is represented by keeping 10 equally spaced MFCCs and ﬂattening these [39]. Both models use unsupervised syllable pre-segmentation [23] to limit allowed word boundaries. ”
Unsupervised End-to-End Learning of Discrete Linguistic Units for Voice Conversion (2019, background): “Embedding audio signals into latent representations has been a well studied practice [4][5][6][7][8][9][10].”
Multi-view Recurrent Neural Acoustic Word Embeddings (2016, methodology): “Levin et al. (2013) proposed an approach for embedding an arbitrary-length segment of speech as a ﬁxed-dimensional vector, based on representing each word as a vector of dynamic time warping (DTW) distances to a set of template words.”
Semantic Query-by-example Speech Search Using Visual Grounding (2019, background): “Even for studies considering unsupervised acoustic embeddings, true word boundaries are normally used (e.g., in [24, 27]).”
Learning Word Embeddings from Speech (2017, background): “Previous research has explored the concept of learning vector representations from speech [He et al., 2017, Kamper et al., 2016, Chung et al., 2016, Settle and Livescu, 2016, Bengio and Heigold, 2014, Levin et al., 2013].”
Self-Supervised Speech Representation Learning: A Review (2022, background): “them as vectors of distances from the target segment to a number of pre-defined “template” segments [199].”
Fully unsupervised small-vocabulary speech recognition using a segmental Bayesian model (2015, methodology): “This approach has found success in other unsupervised studies [19, 20, 21] since search and clustering tasks are more efficient in a fixeddimensional space.”
Unsupervised Word Segmentation and Lexicon Discovery Using Acoustic Word Embeddings (2016, methodology): “Based on [29], [32] we use the following parameters for the fixeddimensional embedding extraction (Section III-A): dimensionality D = 11, k = 30, σK = 0.”
Effectiveness of Self-Supervised Pre-Training for ASR (2020, background): “…for new conditions with limited resources spurred a lot of work focused on unsupervised speech recognition and representation learning [2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14], in addition to semi-and weakly-supervised learning techniques to reduce the supervised data needed in real-world…”
Query-by-Example Search with Discriminative Neural Acoustic Word Embeddings (2017, background): “5 seconds and the orthographic representation of each word type has at least six characters [8, 10, 16].”
Audio Word2vec: Sequence-to-Sequence Autoencoding for Unsupervised Learning of Audio Segmentation and Representation (2019, background): “When based on audio word2vec, QbE STD is much more efficient than conventional dynamic time warping (DTW) based approaches, because only the similarities between two single vectors are needed; this is in addition to the significantly better retrieval performance that it yields.”