Dynamic Encoder Size Based On Data-driven Layer-wise Pruning For Speech Recognition

Abstract

Varying-size models are often required to deploy ASR systems under different hardware and/or application constraints such as memory and latency. To avoid redundant training and optimization efforts for individual models of different sizes, we present the dynamic encoder size approach, which jointly trains multiple performant models within one supernet from scratch. These subnets of various sizes are layer-wise pruned from the supernet, and thus, enjoy full parameter sharing. By combining score-based pruning with supernet training, we propose two novel methods, Simple-Top-k and Iterative-Zero-Out, to automatically select the best-performing subnets in a data-driven manner, avoiding resource-intensive search efforts. Our experiments using CTC on both Librispeech and TED-LIUM-v2 corpora show that our methods can achieve on-par performance as individually trained models of each size category. Also, our approach consistently brings small performance improvements for the full-size supernet.

Stats

• citations4
• S2 citations—
• github stars0
• HF likes0
• heat score5.24
• arxiv keyxu2024dynamic

Related papers

• Accurate And Structured Pruning For Efficient Automatic Speech Recognition (2023)7.81
• Dynamic Sparsity Neural Networks For Automatic Speech Recognition (2020)0.00
• Structured Pruning Of Self-supervised Pre-trained Models For Speech Recognition And Understanding (2023)11.39
• Training Dynamic Models Using Early Exits For Automatic Speech Recognition On Resource-constrained Devices (2023)0.00
• CTC Blank Triggered Dynamic Layer-skipping For Efficient Ctc-based Speech Recognition (2024)0.00
• Extreme Encoder Output Frame Rate Reduction: Improving Computational Latencies Of Large End-to-end Models (2024)5.84
• EPIC TTS Models: Empirical Pruning Investigations Characterizing Text-to-speech Models (2022)3.58
• Federated Pruning: Improving Neural Network Efficiency With Federated Learning (2022)7.50