iflow-mcp_neosun100_glm-asr 1.0.0

GLM-ASR - All-in-One Speech Recognition Service based on GLM-ASR-Nano

English | 简体中文 | 繁體中文 | 日本語

GLM-ASR

All-in-One Speech Recognition Service based on GLM-ASR-Nano

Web UI • REST API • SSE Streaming • Swagger Docs

---

🖥️ Screenshot

---

✨ Features

• 🎯 High Accuracy - Based on GLM-ASR-Nano-2512 (1.5B), outperforms Whisper V3
• 🌍 17 Languages - Chinese, English, Cantonese, Japanese, Korean, and more
• 🎤 Long Audio - VAD smart segmentation for unlimited audio length
• 🚀 SSE Streaming - Real-time progress and results for long audio
• 🖥️ Web UI - Modern dark-mode interface with 4 language support
• 🔌 REST API - Full API with Swagger documentation
• 💾 GPU Management - Manual load/unload for memory control
• 🐳 Docker Ready - One-command deployment with pre-loaded model

---

🚀 Quick Start

Docker (Recommended)

docker run -d --gpus all -p 7860:7860 neosun/glm-asr:v2.0.1

Access:

• Web UI: http://localhost:7860
• Swagger Docs: http://localhost:7860/docs
• ReDoc: http://localhost:7860/redoc

Docker Compose

git clone https://github.com/neosun100/glm-asr.git
cd glm-asr
docker compose up -d

---

📖 API Reference

Base URL

http://localhost:7860

Endpoints

Health Check

GET /health

{"status": "ok", "model_loaded": true}

Transcribe (Sync) - For short audio

POST /api/transcribe
Content-Type: multipart/form-data

Parameter	Type	Default	Description
file	File	required	Audio file (wav/mp3/flac/m4a/ogg/webm)
max_new_tokens	int	512	Max output tokens (1-2048)

curl -X POST http://localhost:7860/api/transcribe \
  -F "file=@audio.mp3" \
  -F "max_new_tokens=512"

{"status": "success", "text": "Transcribed text here..."}

Transcribe (SSE Stream) - For long audio

POST /api/transcribe/stream
Content-Type: multipart/form-data

Returns Server-Sent Events with real-time progress:

Event Type	Description	Example
start	Processing started	{"type": "start"}
progress	Segment progress	{"type": "progress", "current": 3, "total": 10, "duration": 22.5}
partial	Segment result	{"type": "partial", "text": "Segment text..."}
done	Complete	{"type": "done", "text": "Full transcription..."}
error	Error occurred	{"type": "error", "message": "Error details"}

curl -X POST http://localhost:7860/api/transcribe/stream \
  -F "file=@long_audio.mp3"

GPU Status

GET /gpu/status

{
  "model_loaded": true,
  "device": "cuda",
  "gpu_memory_used_mb": 4320.5,
  "gpu_memory_total_mb": 24576.0
}

Load/Unload Model

POST /gpu/load
POST /gpu/unload

Interactive Documentation

• Swagger UI: http://localhost:7860/docs
• ReDoc: http://localhost:7860/redoc

---

⚙️ Configuration

Environment Variables

Variable	Default	Description
MODEL_CHECKPOINT	zai-org/GLM-ASR-Nano-2512	HuggingFace model path
PORT	7860	Service port
HF_HOME	/app/cache	Model cache directory

docker-compose.yml

services:
  glm-asr:
    image: neosun/glm-asr:v2.0.1
    container_name: glm-asr
    ports:
      - "7860:7860"
    volumes:
      - ./cache:/app/cache
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: 1
              capabilities: [gpu]

---

🏗️ Tech Stack

Component	Technology
Model	GLM-ASR-Nano-2512 (1.5B)
Backend	FastAPI + Uvicorn
Streaming	Server-Sent Events (SSE)
Frontend	HTML5 + Vanilla JS
Container	Docker + NVIDIA CUDA
API Docs	Swagger / ReDoc

---

📊 Benchmark

GLM-ASR-Nano achieves the lowest average error rate (4.10) among comparable models:

---

📝 Changelog

v2.0.1 (2025-12-28)

• ✅ Migrated to FastAPI async framework
• ✅ SSE streaming for real-time progress
• ✅ Complete Swagger API documentation
• ✅ Dual API mode: sync + streaming
• ✅ Fixed browser timeout for long audio
• ✅ Modern dark UI with progress display

v1.1.0 (2025-12-15)

• ✅ VAD smart segmentation (silero-vad)
• ✅ Support unlimited audio length

v1.0.0 (2025-12-14)

• ✅ Initial release
• ✅ Web UI with 4 language support
• ✅ REST API with Swagger docs
• ✅ Docker all-in-one image

---

📄 License

Apache License 2.0

---

⭐ Star History