GPU-Accelerated Viterbi Exact Lattice Decoder for Batched Online and Offline Speech Recognition
This enables deployment of production-grade ASR models on a wide range of systems, from data centers to edge devices, by significantly improving decoding speed and efficiency.
The authors tackled the problem of efficient speech recognition decoding by developing a GPU-accelerated WFST decoder that supports both online streaming and offline batch processing, achieving up to a 240x speedup over single-core CPU decoding and up to 40x faster decoding than the current state-of-the-art GPU decoder.
We present an optimized weighted finite-state transducer (WFST) decoder capable of online streaming and offline batch processing of audio using Graphics Processing Units (GPUs). The decoder is efficient in memory utilization, input/output (I/O) bandwidth, and uses a novel Viterbi implementation designed to maximize parallelism. The reduced memory footprint allows the decoder to process significantly larger graphs than previously possible, while optimizing I/O increases the number of simultaneous streams supported. GPU preprocessing of lattice segments enables intermediate lattice results to be returned to the requestor during streaming inference. Collectively, the proposed algorithm yields up to a 240x speedup over single core CPU decoding, and up to 40x faster decoding than the current state-of-the-art GPU decoder, while returning equivalent results. This decoder design enables deployment of production-grade ASR models on a large spectrum of systems, ranging from large data center servers to low-power edge devices.