stdo/PengChengStarling
By stdo · ↓ 4 · ♥ 7 · Updated 2025-02-11
What's in this repo
This page lists the files in stdo/PengChengStarling, including decoder-epoch-75-avg-11-chunk-16-left-128.onnx, encoder-epoch-75-avg-11-chunk-16-left-128.onnx, joiner-epoch-75-avg-11-chunk-16-left-128.onnx.
- 3 ONNX models (.onnx files)
- 15 files, including 7 .txt, 3 .onnx, 1 .md, 1 .model, 1 .json, 1 .pt
- Last updated 2025-02-11
ONNX files
| File | Size | Download |
|---|---|---|
| decoder-epoch-75-avg-11-chunk-16-left-128.onnx | — | Download |
| encoder-epoch-75-avg-11-chunk-16-left-128.onnx | — | Download |
| joiner-epoch-75-avg-11-chunk-16-left-128.onnx | — | Download |
How to use
These ONNX files can be loaded with ONNX Runtime. To inspect a model graph, upload the file to the Netron viewer. Download links point to the original files on Hugging Face.
README
Introduction
The PengChengStarling project is a multilingual ASR system development toolkit built upon the icefall project. To evaluate the capabilities of PengChengStarling, we developed a multilingual streaming ASR model supporting eight languages: Chinese, English, Russian, Vietnamese, Japanese, Thai, Indonesian, and Arabic. Each language was trained with approximately 2,000 hours of audio data, primarily sourced from open datasets. Our model achieves comparable or superior streaming ASR performance in six of these languages compared to Whisper-Large v3, while being only 20% of its size. Additionally, our model offers a remarkable 7x speed improvement in inference compared to Whisper-Large v3.
Results
| Language | Testset | Whisper-Large v3 | Ours |
|---|---|---|---|
| Chinese | wenetspeech test meeting | 22.99 | 22.67 |
| Vietnamese | gigaspeech2-vi test | 17.94 | 7.09 |
| Japanese | reazonspeech test | 16.3 | 13.34 |
| Thai | gigaspeech2-th test | 20.44 | 17.39 |
| Indonesia | gigaspeech2-id test | 20.03 | 20.54 |
| Arabic | mgb2 test | 30.3 | 24.37 |
Uses
Please refer to the document for guidance on using the checkpoints in this repository.
Model Card Contact
yangb05@pcl.ac.cn
Model info
Created: 2025-01-06
Updated: 2025-02-11
View on Hugging FaceFiles (15)
Related models
More from stdo
No other models by this author.
More models
No task type declared.