A hybrid dynamic time warping-deep neural network architecture for unsupervised acoustic modeling

By Roland Thiollière, Ewan Dunbar, Gabriel Synnaeve, Maarten Versteegh, Emmanuel Dupoux. 2015

75 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Model Architectures, 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 Model Architectures

In Speech & Audio

In Signal Processing

Citations in the corpus

Learning Hierarchical Discrete Linguistic Units from Visually-Grounded Speech (2019, methodology): “Deep neural network models have been successfully used to learn powerful speech representations using weakly or unsupervised objectives (Thiolliere et al., 2015; Kamper et al., 2015; Hsu et al., 2017a;b; Hsu & Glass, 2018; Holzenberger et al., 2018; Milde & Biemann, 2018; van den Oord et al., 2018;…”
Unsupervised acoustic unit discovery for speech synthesis using discrete latent-variable neural networks (2019, neutral): “As part of the 2015 and 2017 Zero Resource Speech Challenges (ZRSC), several unsupervised methods were proposed for learning continuous acoustic representations [6–11].”
The zero resource speech challenge 2017 (2017, background): “A Siamese network [27] might improve the result.”
A segmental framework for fully-unsupervised large-vocabulary speech recognition (2016, methodology): “In this paper we show that cAE-learned features also improve performance of our multi-speaker unsupervised segmentation and clustering system.”
Jointly Discovering Visual Objects and Spoken Words from Raw Sensory Input (2018, methodology): “Neural networks have thus far been mostly utilized in this realm for learning frame-level acoustic features (Kamper et al. 2015; Renshaw et al. 2015; Thiolliere et al. 2015; Zhang et al. 2012).”
The Zero Resource Speech Challenge 2015: Proposed Approaches and Results (2016, methodology): “Or the feedback from Track 2 into Track 1 as was shown in several systems [9, 10] could be applied to other feature extraction systems.”
Combining Adversarial Training and Disentangled Speech Representation for Robust Zero-Resource Subword Modeling (2019, background): “In the previous two challenges, several unsupervised representation learning approaches were proposed for comparison, such as cluster posteriorgrams (PGs) [4–6], DNN bottle-neck features [7,8], autoencoders (AEs) [9,10], variational AEs (VAEs) [11,12] and siamese networks [13–15].”
Towards Bilingual Lexicon Discovery From Visually Grounded Speech Audio (2019, methodology): “Deep learning has been used to learn robust feature representation for speech over varying speaker and background characteristics such as in [13, 14, 15, 16].”
Speaker Invariant Feature Extraction for Zero-Resource Languages with Adversarial Learning (2018, methodology): “We therefore plan to apply the proposed multi-task learning to various kinds of neural networks such a bottleneck approach based on multi-lingual phonemes [23] and a Siamese network [7].”
Deep convolutional acoustic word embeddings using word-pair side information (2015, background): “During testing, some inputs may correspond to unseen words, but even in these cases the softmax layer gives a fixed-dimensional distributional representation of the input in terms of seen word types.”
ABX-Discriminability Measures and Applications. (Mesures de discriminabilité ABX et applications) (2016, methodology): “The Siamese Networks approach from [97] was tested [101] and compared to an approach [102] also training an NN-based acoustic model but using Correspondence Autoencoders.”
Deep learning methods for unsupervised acoustic modeling — Leap submission to ZeroSpeech challenge 2017 (2017, unknown): “A hybrid dynamic time warping (DTW) approach has also been previously attempted for the zero resource challenge [10].”
Feature optimized DPGMM clustering for unsupervised subword modeling: A contribution to zerospeech 2017 (2017, background): “[5] propose a Siamese DNN training framework that takes the frames of UTD word pairs as input and minimizes the distance between frames of the same class and maximizes it between frames of different classes.”
Learning Weakly Supervised Multimodal Phoneme Embeddings (2017, methodology): “In particular, we will use a Siamese DNN architecture which feeds on word-level side information (the fact that two words are the same or different: [5, 6, 7, 8]), which demonstrably can be discovered automatically from continuous speech using spoFigure 1: Multi-task setting.”