Unbiased Semi-Supervised LF-MMI Training Using Dropout

Tong, Sibo; Vyas, Apoorv; Garner, Philip N.; Bourlard, Hervé

doi:10.21437/interspeech.2019-2678

Cited by 6 publications

(7 citation statements)

References 14 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Based on this observation, we devised a novel framework which uses Dropout at the test time to sample from the posterior predictive distribution of word-sequences to produce unbiased supervision for semi-supervised training. Results on monolingual experiments show that the proposed approach can further improve the performance over the state-of-the-art method [Tong et al, 2019b].…”

Section: Main Contributionsmentioning

confidence: 99%

See 1 more Smart Citation

Multi-Spectral Widefield Microscopy of the Beating Heart Through Post-Acquisition Synchronization and Unmixing

Jaques

Bapst-Wicht

Liebling

2019

2019 IEEE 16th International Symposium on Biomedical Imaging (ISBI 2019)

View full text Add to dashboard Cite

State-of-the-art acoustic models for Automatic Speech Recognition (ASR) are based on Hidden Markov Models (HMM) and Deep Neural Networks (DNN) and often require thousands of hours of transcribed speech data during training. Therefore, building multilingual ASR systems or systems on a language with few resources is a challenging task. Multilingual training and cross-lingual adaptation are potential solutions. However, context-dependent states modeling creates difficulties for multilingual and cross-lingual ASR because of the large increase in context dependent labels arising from the phone set mismatch.

show abstract

Section: Main Contributionsmentioning

confidence: 99%

“…Therefore, we present a novel approach for semi-supervised training in this chapter, which can address the data scarcity problem from a different aspect. The work in this chapter was published as Tong et al [2019b].…”

Section: Semi-supervised Training Using Dropoutmentioning

confidence: 99%

Multi-Spectral Widefield Microscopy of the Beating Heart Through Post-Acquisition Synchronization and Unmixing

Jaques

Bapst-Wicht

Liebling

2019

2019 IEEE 16th International Symposium on Biomedical Imaging (ISBI 2019)

View full text Add to dashboard Cite

show abstract

“…A typical approach to exploit unsupervised data for automatic speech recognition (ASR) is to train a seed model using supervised data and use the seed model to automatically transcribe the unsupervised data [1,2,3,4,5,6,7,8,9]. This approach is referred to as semi-supervised learning.…”

Section: Introductionmentioning

confidence: 99%

Lattice-Free Mmi Adaptation of Self-Supervised Pretrained Acoustic Models

Vyas

Madikeri

Bourlard

2021

ICASSP 2021 - 2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)

Self Cite

View full text Add to dashboard Cite

In this work, we propose lattice-free MMI (LFMMI) for supervised adaptation of self-supervised pretrained acoustic model. We pretrain a Transformer model on thousand hours of untranscribed Librispeech data followed by supervised adaptation with LFMMI on three different datasets. Our results show that fine-tuning with LFMMI, we consistently obtain relative WER improvements of 10% and 35.3% on the clean and other test sets of Librispeech (100h), 10.8% on Switchboard (300h), and 4.3% on Swahili (38h) and 4.4% on Tagalog (84h) compared to the baseline trained only with supervised data.

show abstract

“…In the aforementioned approaches, the entire unlabelled data is decoded only once with a seed model trained using manually labeled data. In [14,15], multiple systems (or outputs) are used to obtain better labels. In [16], interleaved training by continuously updating the model used to generate labels was shown to be effective.…”

Section: Introductionmentioning

confidence: 99%

Incremental Semi-Supervised Learning for Multi-Genre Speech Recognition

Khonglah

Madikeri

Dey

et al. 2020

ICASSP 2020 - 2020 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)

View full text Add to dashboard Cite

In this work, we explore a data scheduling strategy for semisupervised learning (SSL) for acoustic modeling in automatic speech recognition. The conventional approach uses a seed model trained with supervised data to automatically recognize the entire set of unlabeled (auxiliary) data to generate new labels for subsequent acoustic model training. In this paper, we propose an approach in which the unlabelled set is divided into multiple equal-sized subsets. These subsets are processed in an incremental fashion: for each iteration a new subset is added to the data used for SSL, starting from only one subset in the first iteration. The acoustic model from the previous iteration becomes the seed model for the next one. This scheduling strategy is compared to the approach employing all unlabeled data in one-shot for training. Experiments using lattice-free maximum mutual information based acoustic model training on Fisher English gives 80% word error recovery rate. On the multi-genre evaluation sets on Lithuanian and Bulgarian relative improvements of up to 17.2% in word error rate are observed.

show abstract

Unbiased Semi-Supervised LF-MMI Training Using Dropout

Cited by 6 publications

References 14 publications

Multi-Spectral Widefield Microscopy of the Beating Heart Through Post-Acquisition Synchronization and Unmixing

Multi-Spectral Widefield Microscopy of the Beating Heart Through Post-Acquisition Synchronization and Unmixing

Lattice-Free Mmi Adaptation of Self-Supervised Pretrained Acoustic Models

Incremental Semi-Supervised Learning for Multi-Genre Speech Recognition

Contact Info

Product

Resources

About