• DOI:10.48550/arXiv.2211.16270
• Corpus ID: 254069712

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

@article{Braun2022NeuralTT,
  title={Neural Transducer Training: Reduced Memory Consumption with Sample-wise Computation},
  author={Stefan Braun and Erik McDermott and Roger Hsiao},
  journal={ArXiv},
  year={2022},
  volume={abs/2211.16270}
}

• Stefan Braun, E. McDermott, Roger Hsiao
• Published 29 November 2022
• Computer Science
• ArXiv

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.