timmylucy/GLM-ASR
🔊 Enhance speech recognition with GLM-ASR-Nano-2512, a high-performance model excelling in dialect support and low-volume audio accuracy.
What's novel
🔊 Enhance speech recognition with GLM-ASR-Nano-2512, a high-performance model excelling in dialect support and low-volume audio accuracy.
Code Analysis
6 files read · 3 roundsA single ~150-line CLI inference script that loads a pre-trained 1.5B ASR model from HuggingFace, chunks audio into 30-second segments, extracts Whisper-style mel features, and generates a transcript via a causal LM.
Strengths
The inference script is functional and correctly implements the audio chunking, feature extraction, and token-sequence construction pipeline for the GLM-ASR model. The Chinese README is honest about what the repo contains and provides clear usage instructions.
Weaknesses
The English README is severely misleading, describing a GUI application with installers and record buttons that do not exist in the code. There are zero tests, minimal error handling, a use of eval() on a hardcoded string, and no library/API structure — it is purely a demo script with no reusable architecture.
Score Breakdown
Signal breakdown
Innovation
Craft
Traction
Scope
Evidence
Commits
14
Contributors
4
Files
16
Active weeks
3
Repository
Language
Python
Stars
1
Forks
0
License
Apache-2.0