fixed-point quantization

Kuaishou Tech

Nov 8, 2021 · Artificial Intelligence

FPGA-Based Real-Time Streaming ASR Acceleration for Kuaishou: A Case Study in Domain-Specific Hardware Optimization

This paper presents a full fixed-point FPGA-based hardware acceleration solution for TDNN+LSTM acoustic models in real-time streaming ASR, achieving 37.67% latency reduction and 7.5x concurrency improvement through software-hardware co-design and domain-specific optimization.