Get our free extension to see links to code for papers anywhere online!

Chrome logo Add to Chrome

Firefox logo Add to Firefox

"speech": models, code, and papers

Layer Reduction: Accelerating Conformer-Based Self-Supervised Model via Layer Consistency

Apr 08, 2021
Jinchuan Tian, Rongzhi Gu, Helin Wang, Yuexian Zou

Transformer-based self-supervised models are trained as feature extractors and have empowered many downstream speech tasks to achieve state-of-the-art performance. However, both the training and inference process of these models may encounter prohibitively high computational cost and large parameter budget. Although Parameter Sharing Strategy (PSS) proposed in ALBERT paves the way for parameter reduction, the computation required remains the same. Interestingly, we found in experiments that distributions of feature embeddings from different Transformer layers are similar when PSS is integrated: a property termed as Layer Consistency (LC) in this paper. Given this similarity of feature distributions, we assume that feature embeddings from different layers would have similar representing power. In this work, Layer Consistency enables us to adopt Transformer-based models in a more efficient manner: the number of Conformer layers in each training iteration could be uniformly sampled and Shallow Layer Inference (SLI) could be applied to reduce the number of layers in inference stage. In experiments, our models are trained with LibriSpeech dataset and then evaluated on both phone classification and Speech Recognition tasks. We experimentally achieve 7.8X parameter reduction, 41.9% training speedup and 37.7% inference speedup while maintaining comparable performance with conventional BERT-like self-supervised methods.

* 5 pages, 3 figures, submit to Interspeech2021 

  Access Paper or Ask Questions

Multimodal Attention Fusion for Target Speaker Extraction

Feb 02, 2021
Hiroshi Sato, Tsubasa Ochiai, Keisuke Kinoshita, Marc Delcroix, Tomohiro Nakatani, Shoko Araki

Target speaker extraction, which aims at extracting a target speaker's voice from a mixture of voices using audio, visual or locational clues, has received much interest. Recently an audio-visual target speaker extraction has been proposed that extracts target speech by using complementary audio and visual clues. Although audio-visual target speaker extraction offers a more stable performance than single modality methods for simulated data, its adaptation towards realistic situations has not been fully explored as well as evaluations on real recorded mixtures. One of the major issues to handle realistic situations is how to make the system robust to clue corruption because in real recordings both clues may not be equally reliable, e.g. visual clues may be affected by occlusions. In this work, we propose a novel attention mechanism for multi-modal fusion and its training methods that enable to effectively capture the reliability of the clues and weight the more reliable ones. Our proposals improve signal to distortion ratio (SDR) by 1.0 dB over conventional fusion mechanisms on simulated data. Moreover, we also record an audio-visual dataset of simultaneous speech with realistic visual clue corruption and show that audio-visual target speaker extraction with our proposals successfully work on real data.

* in IEEE Spoken Language Technology Workshop (SLT), 2021, pp. 778-784 
* 7 pages, 5 figures 

  Access Paper or Ask Questions

Interpretable Visualization and Higher-Order Dimension Reduction for ECoG Data

Dec 12, 2020
Kelly Geyer, Frederick Campbell, Andersen Chang, John Magnotti, Michael Beauchamp, Genevera I. Allen

ElectroCOrticoGraphy (ECoG) technology measures electrical activity in the human brain via electrodes placed directly on the cortical surface during neurosurgery. Through its capability to record activity at a fast temporal resolution, ECoG experiments have allowed scientists to better understand how the human brain processes speech. By its nature, ECoG data is difficult for neuroscientists to directly interpret for two major reasons. Firstly, ECoG data tends to be large in size, as each individual experiment yields data up to several gigabytes. Secondly, ECoG data has a complex, higher-order nature. After signal processing, this type of data may be organized as a 4-way tensor with dimensions representing trials, electrodes, frequency, and time. In this paper, we develop an interpretable dimension reduction approach called Regularized Higher Order Principal Components Analysis, as well as an extension to Regularized Higher Order Partial Least Squares, that allows neuroscientists to explore and visualize ECoG data. Our approach employs a sparse and functional Candecomp-Parafac (CP) decomposition that incorporates sparsity to select relevant electrodes and frequency bands, as well as smoothness over time and frequency, yielding directly interpretable factors. We demonstrate the performance and interpretability of our method with an ECoG case study on audio and visual processing of human speech.

  Access Paper or Ask Questions

Personalized Speech2Video with 3D Skeleton Regularization and Expressive Body Poses

Jul 17, 2020
Miao Liao, Sibo Zhang, Peng Wang, Hao Zhu, Ruigang Yang

In this paper, we propose a novel approach to convert given speech audio to a photo-realistic speaking video of a specific person, where the output video has synchronized, realistic, and expressive rich body dynamics. We achieve this by first generating 3D skeleton movements from the audio sequence using a recurrent neural network (RNN), and then synthesizing the output video via a conditional generative adversarial network (GAN). To make the skeleton movement realistic and expressive, we embed the knowledge of an articulated 3D human skeleton and a learned dictionary of personal speech iconic gestures into the generation process in both learning and testing pipelines. The former prevents the generation of unreasonable body distortion, while the later helps our model quickly learn meaningful body movement through a few recorded videos. To produce photo-realistic and high-resolution video with motion details, we propose to insert part attention mechanisms in the conditional GAN, where each detailed part, e.g. head and hand, is automatically zoomed in to have their own discriminators. To validate our approach, we collect a dataset with 20 high-quality videos from 1 male and 1 female model reading various documents under different topics. Compared with previous SoTA pipelines handling similar tasks, our approach achieves better results by a user study.

  Access Paper or Ask Questions

Disentangling Style and Speaker Attributes for TTS Style Transfer

Jan 24, 2022
Xiaochun An, Frank K. Soong, Lei Xie

End-to-end neural TTS has shown improved performance in speech style transfer. However, the improvement is still limited by the available training data in both target styles and speakers. Additionally, degenerated performance is observed when the trained TTS tries to transfer the speech to a target style from a new speaker with an unknown, arbitrary style. In this paper, we propose a new approach to seen and unseen style transfer training on disjoint, multi-style datasets, i.e., datasets of different styles are recorded, one individual style by one speaker in multiple utterances. An inverse autoregressive flow (IAF) technique is first introduced to improve the variational inference for learning an expressive style representation. A speaker encoder network is then developed for learning a discriminative speaker embedding, which is jointly trained with the rest neural TTS modules. The proposed approach of seen and unseen style transfer is effectively trained with six specifically-designed objectives: reconstruction loss, adversarial loss, style distortion loss, cycle consistency loss, style classification loss, and speaker classification loss. Experiments demonstrate, both objectively and subjectively, the effectiveness of the proposed approach for seen and unseen style transfer tasks. The performance of our approach is superior to and more robust than those of four other reference systems of prior art.

  Access Paper or Ask Questions

Advancing Momentum Pseudo-Labeling with Conformer and Initialization Strategy

Oct 11, 2021
Yosuke Higuchi, Niko Moritz, Jonathan Le Roux, Takaaki Hori

Pseudo-labeling (PL), a semi-supervised learning (SSL) method where a seed model performs self-training using pseudo-labels generated from untranscribed speech, has been shown to enhance the performance of end-to-end automatic speech recognition (ASR). Our prior work proposed momentum pseudo-labeling (MPL), which performs PL-based SSL via an interaction between online and offline models, inspired by the mean teacher framework. MPL achieves remarkable results on various semi-supervised settings, showing robustness to variations in the amount of data and domain mismatch severity. However, there is further room for improving the seed model used to initialize the MPL training, as it is in general critical for a PL-based method to start training from high-quality pseudo-labels. To this end, we propose to enhance MPL by (1) introducing the Conformer architecture to boost the overall recognition accuracy and (2) exploiting iterative pseudo-labeling with a language model to improve the seed model before applying MPL. The experimental results demonstrate that the proposed approaches effectively improve MPL performance, outperforming other PL-based methods. We also present in-depth investigations to make our improvements effective, e.g., with regard to batch normalization typically used in Conformer and LM quality.

* Submitted to ICASSP2022 

  Access Paper or Ask Questions

Dataset for Identification of Homophobia and Transophobia in Multilingual YouTube Comments

Sep 01, 2021
Bharathi Raja Chakravarthi, Ruba Priyadharshini, Rahul Ponnusamy, Prasanna Kumar Kumaresan, Kayalvizhi Sampath, Durairaj Thenmozhi, Sathiyaraj Thangasamy, Rajendran Nallathambi, John Phillip McCrae

The increased proliferation of abusive content on social media platforms has a negative impact on online users. The dread, dislike, discomfort, or mistrust of lesbian, gay, transgender or bisexual persons is defined as homophobia/transphobia. Homophobic/transphobic speech is a type of offensive language that may be summarized as hate speech directed toward LGBT+ people, and it has been a growing concern in recent years. Online homophobia/transphobia is a severe societal problem that can make online platforms poisonous and unwelcome to LGBT+ people while also attempting to eliminate equality, diversity, and inclusion. We provide a new hierarchical taxonomy for online homophobia and transphobia, as well as an expert-labelled dataset that will allow homophobic/transphobic content to be automatically identified. We educated annotators and supplied them with comprehensive annotation rules because this is a sensitive issue, and we previously discovered that untrained crowdsourcing annotators struggle with diagnosing homophobia due to cultural and other prejudices. The dataset comprises 15,141 annotated multilingual comments. This paper describes the process of building the dataset, qualitative analysis of data, and inter-annotator agreement. In addition, we create baseline models for the dataset. To the best of our knowledge, our dataset is the first such dataset created. Warning: This paper contains explicit statements of homophobia, transphobia, stereotypes which may be distressing to some readers.

* 44 Pages 

  Access Paper or Ask Questions

Confusion2vec 2.0: Enriching Ambiguous Spoken Language Representations with Subwords

Feb 19, 2021
Prashanth Gurunath Shivakumar, Panayiotis Georgiou, Shrikanth Narayanan

Word vector representations enable machines to encode human language for spoken language understanding and processing. Confusion2vec, motivated from human speech production and perception, is a word vector representation which encodes ambiguities present in human spoken language in addition to semantics and syntactic information. Confusion2vec provides a robust spoken language representation by considering inherent human language ambiguities. In this paper, we propose a novel word vector space estimation by unsupervised learning on lattices output by an automatic speech recognition (ASR) system. We encode each word in confusion2vec vector space by its constituent subword character n-grams. We show the subword encoding helps better represent the acoustic perceptual ambiguities in human spoken language via information modeled on lattice structured ASR output. The usefulness of the proposed Confusion2vec representation is evaluated using semantic, syntactic and acoustic analogy and word similarity tasks. We also show the benefits of subword modeling for acoustic ambiguity representation on the task of spoken language intent detection. The results significantly outperform existing word vector representations when evaluated on erroneous ASR outputs. We demonstrate that Confusion2vec subword modeling eliminates the need for retraining/adapting the natural language understanding models on ASR transcripts.

  Access Paper or Ask Questions

Language Through a Prism: A Spectral Approach for Multiscale Language Representations

Nov 09, 2020
Alex Tamkin, Dan Jurafsky, Noah Goodman

Language exhibits structure at different scales, ranging from subwords to words, sentences, paragraphs, and documents. To what extent do deep models capture information at these scales, and can we force them to better capture structure across this hierarchy? We approach this question by focusing on individual neurons, analyzing the behavior of their activations at different timescales. We show that signal processing provides a natural framework for separating structure across scales, enabling us to 1) disentangle scale-specific information in existing embeddings and 2) train models to learn more about particular scales. Concretely, we apply spectral filters to the activations of a neuron across an input, producing filtered embeddings that perform well on part of speech tagging (word-level), dialog speech acts classification (utterance-level), or topic classification (document-level), while performing poorly on the other tasks. We also present a prism layer for training models, which uses spectral filters to constrain different neurons to model structure at different scales. Our proposed BERT + Prism model can better predict masked tokens using long-range context and produces multiscale representations that perform better at utterance- and document-level tasks. Our methods are general and readily applicable to other domains besides language, such as images, audio, and video.

* NeurIPS 2020 

  Access Paper or Ask Questions