Add model card (README.md)
Browse files
README.md
ADDED
|
@@ -0,0 +1,61 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
language:
|
| 3 |
+
- multilingual
|
| 4 |
+
license: cc-by-nc-4.0
|
| 5 |
+
tags:
|
| 6 |
+
- language-identification
|
| 7 |
+
- onnx
|
| 8 |
+
- audio
|
| 9 |
+
- wav2vec2
|
| 10 |
+
- mms-lid
|
| 11 |
+
datasets:
|
| 12 |
+
- mms-lid
|
| 13 |
+
---
|
| 14 |
+
|
| 15 |
+
# MMS-LID 256 (ONNX)
|
| 16 |
+
|
| 17 |
+
ONNX export of **MMS-LID** (Massively Multilingual Speech - Language Identification) for **256 languages**. For on-device or server inference without PyTorch.
|
| 18 |
+
|
| 19 |
+
- **Base model:** [facebook/mms-lid-256](https://huggingface.co/facebook/mms-lid-256)
|
| 20 |
+
- **Format:** ONNX
|
| 21 |
+
- **Languages:** 256 (ISO 639-3)
|
| 22 |
+
|
| 23 |
+
## Contents
|
| 24 |
+
|
| 25 |
+
- ONNX model file(s) for the Wav2Vec2-based LID classifier
|
| 26 |
+
- Label mapping (e.g. `labels.json` or `mms_lid_id2label.json`) for index to language code
|
| 27 |
+
|
| 28 |
+
## Input / Output
|
| 29 |
+
|
| 30 |
+
- **Input:** Raw waveform, 16 kHz mono, 10 seconds (160,000 samples)
|
| 31 |
+
- **Output:** Logits over 256 language classes; `argmax` gives the predicted language index. Map index to ISO 639-3 code using the included labels file.
|
| 32 |
+
|
| 33 |
+
## Usage
|
| 34 |
+
|
| 35 |
+
1. Load the ONNX model with your runtime (e.g. ONNX Runtime, or convert further to Core ML for iOS).
|
| 36 |
+
2. Feed 10 seconds of 16 kHz mono float32 audio.
|
| 37 |
+
3. Take `argmax` of the logits output and look up the language code in the labels file.
|
| 38 |
+
|
| 39 |
+
## Related repos
|
| 40 |
+
|
| 41 |
+
| Languages | Format | Repo |
|
| 42 |
+
|-----------|--------|------|
|
| 43 |
+
| 256 | ONNX | **this repo** |
|
| 44 |
+
| 126 | Core ML | [mms-lid-126-coreml](https://huggingface.co/aoiandroid/mms-lid-126-coreml) |
|
| 45 |
+
| 256 | Core ML | [mms-lid-256-coreml](https://huggingface.co/aoiandroid/mms-lid-256-coreml) |
|
| 46 |
+
| 512 | Core ML | [mms-lid-512-coreml](https://huggingface.co/aoiandroid/mms-lid-512-coreml) |
|
| 47 |
+
|
| 48 |
+
## Citation
|
| 49 |
+
|
| 50 |
+
```bibtex
|
| 51 |
+
@article{pratap2023mms,
|
| 52 |
+
title={Scaling Speech Technology to 1,000+ Languages},
|
| 53 |
+
author={Vineel Pratap and Andros Tjandra and Bowen Shi and Paden Tomasello and Arun Babu and Sayani Kundu and Ali Elkahky and Zhaoheng Ni and Apoorv Vyas and Maryam Fazel-Zarandi and Alexei Baevski and Yossi Adi and Xiaohui Zhang and Wei-Ning Hsu and Alexis Conneau and Michael Auli},
|
| 54 |
+
journal={arXiv},
|
| 55 |
+
year={2023}
|
| 56 |
+
}
|
| 57 |
+
```
|
| 58 |
+
|
| 59 |
+
## License
|
| 60 |
+
|
| 61 |
+
CC-BY-NC-4.0 (inherited from MMS-LID).
|