English

Neural Transducer Training: Reduced Memory Consumption with Sample-wise Computation

Computation and Language 2023-03-14 v2 Sound Audio and Speech Processing

Abstract

The neural transducer is an end-to-end model for automatic speech recognition (ASR). While the model is well-suited for streaming ASR, the training process remains challenging. During training, the memory requirements may quickly exceed the capacity of state-of-the-art GPUs, limiting batch size and sequence lengths. In this work, we analyze the time and space complexity of a typical transducer training setup. We propose a memory-efficient training method that computes the transducer loss and gradients sample by sample. We present optimizations to increase the efficiency and parallelism of the sample-wise method. In a set of thorough benchmarks, we show that our sample-wise method significantly reduces memory usage, and performs at competitive speed when compared to the default batched computation. As a highlight, we manage to compute the transducer loss and gradients for a batch size of 1024, and audio length of 40 seconds, using only 6 GB of memory.

Keywords

Cite

@article{arxiv.2211.16270,
  title  = {Neural Transducer Training: Reduced Memory Consumption with Sample-wise Computation},
  author = {Stefan Braun and Erik McDermott and Roger Hsiao},
  journal= {arXiv preprint arXiv:2211.16270},
  year   = {2023}
}

Comments

5 pages, 4 figures, 1 table, 1 algorithm

R2 v1 2026-06-28T07:16:49.262Z