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

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.

Stats

• citations0
• S2 citations—
• github stars0
• HF likes0
• heat score0.00
• arxiv keybraun2022neural

Related papers

• Conv-transformer Transducer: Low Latency, Low Frame Rate, Streamable End-to-end Speech Recognition (2020)11.08
• Improving RNN Transducer Modeling For End-to-end Speech Recognition (2019)0.00
• Transformer-transducer: End-to-end Speech Recognition With Self-attention (2019)0.00
• Minimum Latency Training Of Sequence Transducers For Streaming End-to-end Speech Recognition (2022)0.00
• Phoneme Based Neural Transducer For Large Vocabulary Speech Recognition (2020)9.59
• Transfer Learning For Speech Recognition On A Budget (2017)14.27
• Developing Real-time Streaming Transformer Transducer For Speech Recognition On Large-scale Dataset (2020)0.00
• Transformer Transducer: One Model Unifying Streaming And Non-streaming Speech Recognition (2020)0.00