Model card: usage with the included code
Browse files
README.md
CHANGED
|
@@ -33,7 +33,28 @@ filter) makes the waveform directly. No separate neural vocoder is needed.
|
|
| 33 |
| `tiny_voca_hindi.pt` | PyTorch checkpoint (state dict + config), after 100k training steps | 38.9 MB |
|
| 34 |
| `config.json` | the model configuration | |
|
| 35 |
|
| 36 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 37 |
|
| 38 |
## Training
|
| 39 |
|
|
@@ -48,3 +69,6 @@ The inference code is **not included in this repository**.
|
|
| 48 |
The weights are released under **CC BY 4.0**, following the training data:
|
| 49 |
- voice data: [Rasa](https://huggingface.co/datasets/ai4bharat/Rasa) by AI4Bharat (CC BY 4.0);
|
| 50 |
- architecture and original code: [tiny-voca](https://huggingface.co/n4ze3m/tiny-voca) (Apache-2.0).
|
|
|
|
|
|
|
|
|
|
|
|
| 33 |
| `tiny_voca_hindi.pt` | PyTorch checkpoint (state dict + config), after 100k training steps | 38.9 MB |
|
| 34 |
| `config.json` | the model configuration | |
|
| 35 |
|
| 36 |
+
## Usage
|
| 37 |
+
|
| 38 |
+
The inference code is included: Python (`speak.py`, `tiny_voca/`, `hindi_text/`) and a C++ engine on ggml with a small
|
| 39 |
+
web page (`cpp/`, `web/`).
|
| 40 |
+
|
| 41 |
+
```bash
|
| 42 |
+
git clone https://huggingface.co/FcbLeo/tiny-voca-hindi
|
| 43 |
+
cd tiny-voca-hindi
|
| 44 |
+
pip install -r requirements.txt
|
| 45 |
+
python speak.py "आज 15 अगस्त है और तापमान 32 डिग्री है।" --style happy --out happy.wav
|
| 46 |
+
```
|
| 47 |
+
|
| 48 |
+
Text is normalised first, so numbers, dates, times, money and units are read as Hindi words.
|
| 49 |
+
|
| 50 |
+
C++ engine (CPU only; CMake downloads ggml v0.29.0 during the first build):
|
| 51 |
+
|
| 52 |
+
```bash
|
| 53 |
+
cpp\build.bat # Windows (Visual Studio 2022 C++ tools)
|
| 54 |
+
sh cpp/build.sh # Linux / macOS
|
| 55 |
+
cpp/build/tts_hi --model tts_hi_q8_0.gguf --text "नमस्ते, आप कैसे हैं?" --style conv --out hello.wav
|
| 56 |
+
cpp/build/tts_hi --model tts_hi_q8_0.gguf --serve 8090 --page web --threads 4 # web page on http://127.0.0.1:8090
|
| 57 |
+
```
|
| 58 |
|
| 59 |
## Training
|
| 60 |
|
|
|
|
| 69 |
The weights are released under **CC BY 4.0**, following the training data:
|
| 70 |
- voice data: [Rasa](https://huggingface.co/datasets/ai4bharat/Rasa) by AI4Bharat (CC BY 4.0);
|
| 71 |
- architecture and original code: [tiny-voca](https://huggingface.co/n4ze3m/tiny-voca) (Apache-2.0).
|
| 72 |
+
|
| 73 |
+
The inference code is released under the **Apache License 2.0** (`LICENSE`, `NOTICE`).
|
| 74 |
+
`cpp/third_party/httplib.h` is [cpp-httplib](https://github.com/yhirose/cpp-httplib) (MIT).
|