Update README.md
Browse files
README.md
CHANGED
|
@@ -7,12 +7,10 @@ tags:
|
|
| 7 |
license: cc-by-nc-sa-4.0
|
| 8 |
library_name: pytorch
|
| 9 |
datasets:
|
| 10 |
-
-
|
| 11 |
-
-
|
| 12 |
-
- VLSP2023
|
| 13 |
-
- vietTTS
|
| 14 |
- UEH
|
| 15 |
-
model_name: ZipVoice-Vietnamese-
|
| 16 |
language: vi
|
| 17 |
---
|
| 18 |
|
|
@@ -20,7 +18,7 @@ language: vi
|
|
| 20 |
This model is only intended for **research purposes**.
|
| 21 |
**Access requests must be made using an institutional, academic, or corporate email**. Requests from public email providers will be denied. We appreciate your understanding.
|
| 22 |
|
| 23 |
-
# 🎙️ ZipVoice-Vietnamese-
|
| 24 |
ZipVoice is a series of fast and high-quality zero-shot TTS models based on flow matching.
|
| 25 |
|
| 26 |
Key features:
|
|
@@ -32,7 +30,7 @@ Key features:
|
|
| 32 |
|
| 33 |
4. Multi-mode: support both single-speaker and dialogue speech generation.
|
| 34 |
|
| 35 |
-
This checkpoint is a compact fine-tuned version of ZipVoice trained on
|
| 36 |
|
| 37 |
🔗 For more fine-tuning and inference experiments, visit: https://github.com/k2-fsa/ZipVoice.
|
| 38 |
|
|
@@ -42,8 +40,8 @@ This checkpoint is a compact fine-tuned version of ZipVoice trained on 150 hours
|
|
| 42 |
|
| 43 |
## 📌 Model Details
|
| 44 |
|
| 45 |
-
- **Dataset:**
|
| 46 |
-
- **Total dataset durations:**
|
| 47 |
- **Data processing Technique:**
|
| 48 |
- Remove all music background from audios, using facebook demucs model: https://github.com/facebookresearch/demucs
|
| 49 |
- Do not use audio files shorter than 1 second or longer than 30 seconds.
|
|
@@ -53,7 +51,7 @@ This checkpoint is a compact fine-tuned version of ZipVoice trained on 150 hours
|
|
| 53 |
- **Base Model:** ZipVoice with espeak-ng vi for tokenizer
|
| 54 |
- **GPU:** RTX 3090
|
| 55 |
- **Batch Siz:** Max duration 200
|
| 56 |
-
- **Training Progress:** Stopped at **
|
| 57 |
|
| 58 |
---
|
| 59 |
|
|
|
|
| 7 |
license: cc-by-nc-sa-4.0
|
| 8 |
library_name: pytorch
|
| 9 |
datasets:
|
| 10 |
+
- PhoAudioBook
|
| 11 |
+
- ViVoice
|
|
|
|
|
|
|
| 12 |
- UEH
|
| 13 |
+
model_name: ZipVoice-Vietnamese-2500h
|
| 14 |
language: vi
|
| 15 |
---
|
| 16 |
|
|
|
|
| 18 |
This model is only intended for **research purposes**.
|
| 19 |
**Access requests must be made using an institutional, academic, or corporate email**. Requests from public email providers will be denied. We appreciate your understanding.
|
| 20 |
|
| 21 |
+
# 🎙️ ZipVoice-Vietnamese-2500h
|
| 22 |
ZipVoice is a series of fast and high-quality zero-shot TTS models based on flow matching.
|
| 23 |
|
| 24 |
Key features:
|
|
|
|
| 30 |
|
| 31 |
4. Multi-mode: support both single-speaker and dialogue speech generation.
|
| 32 |
|
| 33 |
+
This checkpoint is a compact fine-tuned version of ZipVoice trained on 2500 hours of Vietnamese speech.
|
| 34 |
|
| 35 |
🔗 For more fine-tuning and inference experiments, visit: https://github.com/k2-fsa/ZipVoice.
|
| 36 |
|
|
|
|
| 40 |
|
| 41 |
## 📌 Model Details
|
| 42 |
|
| 43 |
+
- **Dataset:** PhoAudioBook, ViVoice, TeacherDinh-UEH.
|
| 44 |
+
- **Total dataset durations:** 2500 hours
|
| 45 |
- **Data processing Technique:**
|
| 46 |
- Remove all music background from audios, using facebook demucs model: https://github.com/facebookresearch/demucs
|
| 47 |
- Do not use audio files shorter than 1 second or longer than 30 seconds.
|
|
|
|
| 51 |
- **Base Model:** ZipVoice with espeak-ng vi for tokenizer
|
| 52 |
- **GPU:** RTX 3090
|
| 53 |
- **Batch Siz:** Max duration 200
|
| 54 |
+
- **Training Progress:** Stopped at **525,000 steps at epoch 11**
|
| 55 |
|
| 56 |
---
|
| 57 |
|