Documents
Presentation Slides
Multilingual and Fully Non-Autoregressive ASR with Large Language Model Fusion: A Comprehensive Study
- DOI:
- 10.60864/41dz-se35
- Citation Author(s):
- Submitted by:
- Tongzhou Chen
- Last updated:
- 6 June 2024 - 10:22am
- Document Type:
- Presentation Slides
- Document Year:
- 2024
- Event:
- Presenters:
- Tongzhou Chen
- Paper Code:
- SS-L22.3
- Categories:
- Log in to post comments
In the era of large models, the autoregressive nature of decoding often results in latency serving as a significant bottleneck. We propose a non-autoregressive LM-fused ASR system that effectively leverages the parallelization capabilities of accelerator hardware. Our approach combines the Universal Speech Model (USM) and the PaLM 2 language model in per-segment scoring mode, achieving an average relative WER improvement across all languages of 10.8% on FLEURS and 3.3% on YouTube captioning. Furthermore, our comprehensive ablation study analyzes key parameters such as LLM size, context length, vocabulary size, fusion methodology. For instance, we explore the impact of LLM size ranging from 128M to 340B parameters on ASR performance. This study provides valuable insights into the factors influencing the effectiveness of practical large-scale LM-fused speech recognition systems.