MSRS: Training Multimodal Speech Recognition Models from Scratch with Sparse Mask Optimization

• URL: http://arxiv.org/abs/2406.17614v1
• Date: Tue, 25 Jun 2024 15:00:43 GMT
• Title: MSRS: Training Multimodal Speech Recognition Models from Scratch with Sparse Mask Optimization
• Authors: Adriana Fernandez-Lopez, Honglie Chen, Pingchuan Ma, Lu Yin, Qiao Xiao, Stavros Petridis, Shiwei Liu, Maja Pantic,
• Abstract summary: MSRS achieves competitive results in VSR and AVSR with 21.1% and 0.9% WER on the LRS3 benchmark, while reducing training time by at least 2x. We explore other sparse approaches and show that only MSRS enables training from scratch by implicitly masking the weights affected by vanishing gradients.
• Score: 49.00754561435518
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Pre-trained models have been a foundational approach in speech recognition, albeit with associated additional costs. In this study, we propose a regularization technique that facilitates the training of visual and audio-visual speech recognition models (VSR and AVSR) from scratch. This approach, abbreviated as \textbf{MSRS} (Multimodal Speech Recognition from Scratch), introduces a sparse regularization that rapidly learns sparse structures within the dense model at the very beginning of training, which receives healthier gradient flow than the dense equivalent. Once the sparse mask stabilizes, our method allows transitioning to a dense model or keeping a sparse model by updating non-zero values. MSRS achieves competitive results in VSR and AVSR with 21.1% and 0.9% WER on the LRS3 benchmark, while reducing training time by at least 2x. We explore other sparse approaches and show that only MSRS enables training from scratch by implicitly masking the weights affected by vanishing gradients.

Related papers

• Unified Speech Recognition: A Single Model for Auditory, Visual, and Audiovisual Inputs [73.74375912785689]
  This paper proposes unified training strategies for speech recognition systems. We demonstrate that training a single model for all three tasks enhances VSR and AVSR performance. We also introduce a greedy pseudo-labelling approach to more effectively leverage unlabelled samples.
  arXiv  Detail & Related papers  (2024-11-04T16:46:53Z)
• Full-Rank No More: Low-Rank Weight Training for Modern Speech Recognition Models [46.87216968390808]
  This paper investigates the under-explored area of low-rank weight training for large-scale Conformer-based speech recognition models from scratch. Applying a low-rank structure exclusively to the attention modules can unexpectedly enhance performance. Feed-forward layers present greater challenges, as they begin to exhibit performance degradation with a moderate 50% rank reduction.
  arXiv  Detail & Related papers  (2024-10-10T09:58:35Z)
• Representation Alignment for Generation: Training Diffusion Transformers Is Easier Than You Think [72.48325960659822]
  One main bottleneck in training large-scale diffusion models for generation lies in effectively learning these representations. We study this by introducing a straightforward regularization called REPresentation Alignment (REPA), which aligns the projections of noisy input hidden states in denoising networks with clean image representations obtained from external, pretrained visual encoders. The results are striking: our simple strategy yields significant improvements in both training efficiency and generation quality when applied to popular diffusion and flow-based transformers, such as DiTs and SiTs.
  arXiv  Detail & Related papers  (2024-10-09T14:34:53Z)
• Rethinking Classifier Re-Training in Long-Tailed Recognition: A Simple Logits Retargeting Approach [102.0769560460338]
  We develop a simple logits approach (LORT) without the requirement of prior knowledge of the number of samples per class. Our method achieves state-of-the-art performance on various imbalanced datasets, including CIFAR100-LT, ImageNet-LT, and iNaturalist 2018.
  arXiv  Detail & Related papers  (2024-03-01T03:27:08Z)
• Exploring the Integration of Speech Separation and Recognition with Self-Supervised Learning Representation [83.36685075570232]
  This work provides an insightful investigation of speech separation in reverberant and noisy-reverberant scenarios as an ASR front-end. We explore multi-channel separation methods, mask-based beamforming and complex spectral mapping, as well as the best features to use in the ASR back-end model. A proposed integration using TF-GridNet-based complex spectral mapping and WavLM-based SSLR achieves a 2.5% word error rate in reverberant WHAMR! test set.
  arXiv  Detail & Related papers  (2023-07-23T05:39:39Z)
• Auto-AVSR: Audio-Visual Speech Recognition with Automatic Labels [100.43280310123784]
  We investigate the use of automatically-generated transcriptions of unlabelled datasets to increase the training set size. We demonstrate that increasing the size of the training set, a recent trend in the literature, leads to reduced WER despite using noisy transcriptions. The proposed model achieves new state-of-the-art performance on AV-ASR on LRS2 and LRS3.
  arXiv  Detail & Related papers  (2023-03-25T00:37:34Z)
• A Mixture of Expert Based Deep Neural Network for Improved ASR [4.993304210475779]
  MixNet is a novel deep learning architecture for acoustic model in the context of Automatic Speech Recognition (ASR) In natural speech, overlap in distribution across different acoustic classes is inevitable, which leads to inter-class mis-classification. Experiments are conducted on a large vocabulary ASR task which show that the proposed architecture provides 13.6% and 10.0% relative reduction in word error rates.
  arXiv  Detail & Related papers  (2021-12-02T07:26:34Z)
• Streaming end-to-end speech recognition with jointly trained neural feature enhancement [20.86554979122057]
  We present a streaming end-to-end speech recognition model based on Monotonic Chunkwise Attention (MoCha) jointly trained with enhancement layers. We introduce two training strategies: Gradual Application of Enhanced Features (GAEF) and Gradual Reduction of Enhanced Loss (GREL)
  arXiv  Detail & Related papers  (2021-05-04T02:25:41Z)
• Scalable Deep Compressive Sensing [43.92187349325869]
  Most existing deep learning methods train different models for different subsampling ratios, which brings additional hardware burden. We develop a general framework named scalable deep compressive sensing (SDCS) for the scalable sampling and reconstruction (SSR) of all existing end-to-end-trained models. Experimental results show that models with SDCS can achieve SSR without changing their structure while maintaining good performance, and SDCS outperforms other SSR methods.
  arXiv  Detail & Related papers  (2021-01-20T08:42:50Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.