Dynamic Context-Aware Streaming Pretrained Language Model For Inverse Text Normalization
This work addresses the problem of real-time text formatting for speech recognition systems, particularly in low-resource languages, though it appears incremental as it builds on existing pretrained language models with streaming adaptations.
The paper tackles the challenge of streaming inverse text normalization (ITN) for converting spoken ASR outputs to written text, particularly in low-resource Vietnamese scenarios, by proposing a dynamic context-aware pretrained language model that achieves accuracy comparable to non-streaming ITN and surpasses existing streaming models while maintaining low latency.
Inverse Text Normalization (ITN) is crucial for converting spoken Automatic Speech Recognition (ASR) outputs into well-formatted written text, enhancing both readability and usability. Despite its importance, the integration of streaming ITN within streaming ASR remains largely unexplored due to challenges in accuracy, efficiency, and adaptability, particularly in low-resource and limited-context scenarios. In this paper, we introduce a streaming pretrained language model for ITN, leveraging pretrained linguistic representations for improved robustness. To address streaming constraints, we propose Dynamic Context-Aware during training and inference, enabling adaptive chunk size adjustments and the integration of right-context information. Experimental results demonstrate that our method achieves accuracy comparable to non-streaming ITN and surpasses existing streaming ITN models on a Vietnamese dataset, all while maintaining low latency, ensuring seamless integration into ASR systems.