Update model card with ONNX usage
Browse files
README.md
CHANGED
|
@@ -5,16 +5,22 @@ pipeline_tag: text-to-speech
|
|
| 5 |
tags:
|
| 6 |
- kokoro
|
| 7 |
- vietnamese
|
| 8 |
-
- tts
|
| 9 |
- text-to-speech
|
| 10 |
-
-
|
| 11 |
license: apache-2.0
|
| 12 |
-
library_name: kokoro
|
| 13 |
---
|
| 14 |
|
| 15 |
# Kokoro Vietnamese
|
| 16 |
|
| 17 |
-
Vietnamese Kokoro TTS
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 18 |
|
| 19 |
## Install
|
| 20 |
|
|
@@ -24,70 +30,43 @@ cd Kokoro-Vietnamese
|
|
| 24 |
pip install -e .
|
| 25 |
```
|
| 26 |
|
| 27 |
-
|
| 28 |
-
|
| 29 |
-
## Voices
|
| 30 |
-
|
| 31 |
-
Use these names with `voice=...` in Python or `--voice ...` in the CLI.
|
| 32 |
-
|
| 33 |
-
| Voice | Name |
|
| 34 |
-
| --- | --- |
|
| 35 |
-
| `diem_trinh` | Diễm Trinh |
|
| 36 |
-
| `hung_thinh` | Hưng Thịnh |
|
| 37 |
-
| `mai_linh` | Mai Linh |
|
| 38 |
-
| `mai_loan` | Mai Loan |
|
| 39 |
-
| `manh_dung` | Mạnh Dũng |
|
| 40 |
-
| `my_yen` | Mỹ Yến |
|
| 41 |
-
| `ngoc_huyen` | Ngọc Huyền |
|
| 42 |
-
| `phat_tai` | Phát Tài |
|
| 43 |
-
| `thanh_dat` | Thành Đạt |
|
| 44 |
-
| `thuc_trinh` | Thục Trinh |
|
| 45 |
-
| `tuan_ngoc` | Tuấn Ngọc |
|
| 46 |
-
| `storyvert` | storyvert |
|
| 47 |
-
| `duc_an` | Đức An |
|
| 48 |
-
| `duc_duy` | đức duy |
|
| 49 |
-
|
| 50 |
-
## Python API
|
| 51 |
-
|
| 52 |
-
```python
|
| 53 |
-
import soundfile as sf
|
| 54 |
-
from kokoro_vietnamese import KokoroVietnamese
|
| 55 |
-
|
| 56 |
-
tts = KokoroVietnamese(device="cuda", voice="diem_trinh")
|
| 57 |
-
|
| 58 |
-
audio, phonemes = tts.synthesize(
|
| 59 |
-
"Giữa một buổi chiều yên tĩnh, cô ấy kể lại câu chuyện bằng một giọng nói ấm áp và chậm rãi."
|
| 60 |
-
)
|
| 61 |
|
| 62 |
-
|
| 63 |
-
|
| 64 |
-
```
|
| 65 |
-
|
| 66 |
-
Use another voice:
|
| 67 |
-
|
| 68 |
-
```python
|
| 69 |
-
tts = KokoroVietnamese(device="cuda", voice="mai_linh")
|
| 70 |
-
audio, phonemes = tts.synthesize("Hôm nay trời trong xanh, gió thổi nhẹ qua hiên nhà.")
|
| 71 |
```
|
| 72 |
|
| 73 |
-
##
|
| 74 |
|
| 75 |
```bash
|
| 76 |
kokoro-vietnamese \
|
| 77 |
-
--text "
|
|
|
|
| 78 |
--voice diem_trinh \
|
| 79 |
-
--output sample.wav \
|
| 80 |
--device cuda
|
| 81 |
```
|
| 82 |
|
| 83 |
-
|
| 84 |
|
| 85 |
```bash
|
| 86 |
-
kokoro-vietnamese
|
|
|
|
|
|
|
|
|
|
|
|
|
| 87 |
```
|
| 88 |
|
| 89 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 90 |
|
| 91 |
-
|
| 92 |
-
|
| 93 |
-
- The voicepacks are derived from the LarVoice multi-speaker training set and are distributed as inference artifacts.
|
|
|
|
| 5 |
tags:
|
| 6 |
- kokoro
|
| 7 |
- vietnamese
|
|
|
|
| 8 |
- text-to-speech
|
| 9 |
+
- onnx
|
| 10 |
license: apache-2.0
|
|
|
|
| 11 |
---
|
| 12 |
|
| 13 |
# Kokoro Vietnamese
|
| 14 |
|
| 15 |
+
Fine-tuned Vietnamese Kokoro TTS artifacts.
|
| 16 |
+
|
| 17 |
+
## Files
|
| 18 |
+
|
| 19 |
+
- `kokoro_vi.pth`: PyTorch Kokoro `KModel` checkpoint for inference.
|
| 20 |
+
- `kokoro_vi.onnx`: ONNX Runtime export of the acoustic model.
|
| 21 |
+
- `kokoro_vi_voicepack.pt`: default Vietnamese voicepack.
|
| 22 |
+
- `config.json`: Kokoro config/vocab used by both PyTorch and ONNX inference.
|
| 23 |
+
- `voicepacks/*.pt`: additional Vietnamese voicepacks.
|
| 24 |
|
| 25 |
## Install
|
| 26 |
|
|
|
|
| 30 |
pip install -e .
|
| 31 |
```
|
| 32 |
|
| 33 |
+
For ONNX Runtime:
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 34 |
|
| 35 |
+
```bash
|
| 36 |
+
pip install -e ".[onnx]"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 37 |
```
|
| 38 |
|
| 39 |
+
## PyTorch Inference
|
| 40 |
|
| 41 |
```bash
|
| 42 |
kokoro-vietnamese \
|
| 43 |
+
--text "Xin chào, hôm nay tôi đang kiểm tra giọng đọc tiếng Việt." \
|
| 44 |
+
--output outputs/sample.wav \
|
| 45 |
--voice diem_trinh \
|
|
|
|
| 46 |
--device cuda
|
| 47 |
```
|
| 48 |
|
| 49 |
+
## ONNX Runtime Inference
|
| 50 |
|
| 51 |
```bash
|
| 52 |
+
kokoro-vietnamese-onnx \
|
| 53 |
+
--text "Tường nhà khách đã được sơn lại." \
|
| 54 |
+
--output outputs/onnx.wav \
|
| 55 |
+
--device cpu \
|
| 56 |
+
--print-phonemes
|
| 57 |
```
|
| 58 |
|
| 59 |
+
The ONNX CLI downloads `kokoro_vi.onnx`, `kokoro_vi_voicepack.pt`, and
|
| 60 |
+
`config.json` from this repository when local paths are not provided. Install
|
| 61 |
+
`onnxruntime-gpu` and pass `--device cuda` to use CUDAExecutionProvider when
|
| 62 |
+
available.
|
| 63 |
+
|
| 64 |
+
## Export ONNX Yourself
|
| 65 |
+
|
| 66 |
+
```bash
|
| 67 |
+
kokoro-vietnamese-export-onnx \
|
| 68 |
+
--output outputs/kokoro_vi.onnx
|
| 69 |
+
```
|
| 70 |
|
| 71 |
+
Vietnamese G2P is handled by `vig2p`, matching the GitHub inference and
|
| 72 |
+
training code.
|
|
|