A Learnable, Pinyin-Aware Post-ASR Correction Method with Accent Routing for Domain-Specific Chinese Speech Recognition-Herbal_Audio_Dataset

This project introduces the Hierarchical Phonetic Fuzzy Search (HPFS) framework, a robust post-processing system designed to enhance Automatic Speech Recognition (ASR) performance in domain-specific and multi-accented Chinese environments (e.g., Traditional Chinese Medicine). The framework features: Learnable Pinyin Distance: A metric learning approach to optimize phonetic similarity weights for initials, finals, and tones. Hierarchical Indexing: Efficient candidate retrieval using a combination of Trie-based prefix filtering and BK-tree metric search. Accent Routing: An intelligent workflow that classifies accents and routes audio to specialized correction engines. This repository contains the complete implementation code for training, indexing, and evaluating the HPFS framework, including baseline comparisons and ablation studies. This section contains the audio dataset for this project.