🎤

Oral English Auto-Scoring

Automated oral exam scoring system based on ASR + Rule Engine + LLM, performing speech transcription, feature extraction, and intelligent scoring.

PythonVolcengine ASRDoubao LLMSQLitePandas

Core Work

Designed multi-layer scoring pipeline: ASR recognition → volume/crosstalk filtering → LLM semantic extraction → weighted keyword matching → rule engine scoring

Implemented dual-ASR comparison (with/without context prompts), calculating scoring confidence based on word-level time alignment, auto-flagging low-confidence samples for human review

Introduced LLM-assisted semantic relevance judgment to handle edge cases that pure rule matching cannot cover

Supports batch concurrent evaluation (ThreadPoolExecutor), SQLite multi-level caching (ASR/LLM), JSONL checkpoint resumption

Rule-configurable design (keyword weights, penalty words, thresholds) — adding new question types only requires YAML changes, no code modifications

Tech Stack

Python / Volcengine ASR / Doubao LLM / SQLite / Pandas

🏗️ Architecture diagram coming soon...

📈 Performance metrics coming soon...