init
Browse files
README.md
CHANGED
|
@@ -23,13 +23,17 @@ tags:
|
|
| 23 |
</p>
|
| 24 |
<p>
|
| 25 |
<a href="#introduction">Introduction</a> ·
|
|
|
|
| 26 |
<a href="#installation">Installation</a> ·
|
| 27 |
<a href="#inference">Inference</a> ·
|
|
|
|
| 28 |
<a href="#citation">Citation</a>
|
| 29 |
</p>
|
| 30 |
</div>
|
| 31 |
|
| 32 |
-
|
|
|
|
|
|
|
| 33 |
|
| 34 |
SemBridge bridges discrete semantic supervision and continuous acoustic generation.
|
| 35 |
A shared semantic tokenizer anchors both acoustic latents and autoregressive language-model
|
|
@@ -41,12 +45,25 @@ The unified model supports **text-to-speech**, **note-conditioned singing**, and
|
|
| 41 |
<img src="https://huggingface.co/ASLP-lab/SemBridge/resolve/main/asset/sembridge_model.png" alt="SemBridge architecture: semantic anchoring for continuous-latent autoregressive generation" width="100%">
|
| 42 |
</div>
|
| 43 |
|
| 44 |
-
|
|
|
|
|
|
|
| 45 |
|
| 46 |
- **[2026-10]**: Released the [inference code](https://github.com/ASLP-lab/SemBridge) and [model checkpoints](https://huggingface.co/ASLP-lab/SemBridge).
|
| 47 |
- **[2026-08]**: Released the [arXiv paper](https://arxiv.org/pdf/2608.07462).
|
| 48 |
|
| 49 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 50 |
|
| 51 |
Use **Python 3.10**.
|
| 52 |
|
|
@@ -63,7 +80,9 @@ python -m pip install torch==2.6.0 torchaudio==2.6.0 \
|
|
| 63 |
python -m pip install -e .
|
| 64 |
```
|
| 65 |
|
| 66 |
-
|
|
|
|
|
|
|
| 67 |
|
| 68 |
```python
|
| 69 |
from sembridge import SemBridge
|
|
@@ -85,7 +104,9 @@ model.generate(
|
|
| 85 |
| **SVS** | Lyrics, MIDI pitches, note durations | BigVGAN-Large |
|
| 86 |
| **Unified** | Ordered speech and singing segments | BigVGAN-Large |
|
| 87 |
|
| 88 |
-
|
|
|
|
|
|
|
| 89 |
|
| 90 |
The repository includes **10 complete examples from our [Demo page](https://tiamojames.github.io/SemBridge_Demo/)**:
|
| 91 |
4 TTS, 4 SVS, and 2 Unified cases, with reference audio and prepared requests.
|
|
@@ -110,7 +131,9 @@ Audio is saved to `outputs/examples/<example_id>.wav`. Set `MODEL=/path/to/SemBr
|
|
| 110 |
to use local weights. See the **[example catalog](https://github.com/ASLP-lab/SemBridge/blob/main/examples/README.md)** for the
|
| 111 |
complete list, reference recordings, and guidance on using your own songs.
|
| 112 |
|
| 113 |
-
|
|
|
|
|
|
|
| 114 |
|
| 115 |
```bash
|
| 116 |
# Text-to-speech
|
|
@@ -142,14 +165,24 @@ For SVS, each note contains `text`, MIDI `pitch`, and `duration` **in seconds**.
|
|
| 142 |
See the **[inference guide](https://github.com/ASLP-lab/SemBridge/blob/main/docs/INFERENCE.md)** for reference voices, singing and
|
| 143 |
mixed-generation examples, sampling controls, and offline loading.
|
| 144 |
|
| 145 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 146 |
|
| 147 |
We thank the authors of the following projects for their open research and resources:
|
| 148 |
|
| 149 |
- **Speech generation:** [VoxCPM](https://github.com/OpenBMB/VoxCPM) and [DiTAR](https://arxiv.org/abs/2502.03930).
|
| 150 |
- **Audio codecs:** [DAC](https://github.com/descriptinc/descript-audio-codec), [SAC](https://github.com/Soul-AILab/SAC), and [BigVGAN](https://github.com/NVIDIA/BigVGAN).
|
| 151 |
|
| 152 |
-
|
|
|
|
|
|
|
| 153 |
|
| 154 |
If you find SemBridge useful in your research, please cite:
|
| 155 |
|
|
@@ -166,7 +199,9 @@ If you find SemBridge useful in your research, please cite:
|
|
| 166 |
}
|
| 167 |
```
|
| 168 |
|
| 169 |
-
|
|
|
|
|
|
|
| 170 |
|
| 171 |
SemBridge code is released under Apache-2.0; upstream portions retain their
|
| 172 |
original licenses. See [LICENSE](https://github.com/ASLP-lab/SemBridge/blob/main/LICENSE) for terms and third-party notices.
|
|
|
|
| 23 |
</p>
|
| 24 |
<p>
|
| 25 |
<a href="#introduction">Introduction</a> ·
|
| 26 |
+
<a href="#roadmap">Roadmap</a> ·
|
| 27 |
<a href="#installation">Installation</a> ·
|
| 28 |
<a href="#inference">Inference</a> ·
|
| 29 |
+
<a href="#community">Community</a> ·
|
| 30 |
<a href="#citation">Citation</a>
|
| 31 |
</p>
|
| 32 |
</div>
|
| 33 |
|
| 34 |
+
<a id="introduction"></a>
|
| 35 |
+
|
| 36 |
+
## π Introduction
|
| 37 |
|
| 38 |
SemBridge bridges discrete semantic supervision and continuous acoustic generation.
|
| 39 |
A shared semantic tokenizer anchors both acoustic latents and autoregressive language-model
|
|
|
|
| 45 |
<img src="https://huggingface.co/ASLP-lab/SemBridge/resolve/main/asset/sembridge_model.png" alt="SemBridge architecture: semantic anchoring for continuous-latent autoregressive generation" width="100%">
|
| 46 |
</div>
|
| 47 |
|
| 48 |
+
<a id="news"></a>
|
| 49 |
+
|
| 50 |
+
## π£ News
|
| 51 |
|
| 52 |
- **[2026-10]**: Released the [inference code](https://github.com/ASLP-lab/SemBridge) and [model checkpoints](https://huggingface.co/ASLP-lab/SemBridge).
|
| 53 |
- **[2026-08]**: Released the [arXiv paper](https://arxiv.org/pdf/2608.07462).
|
| 54 |
|
| 55 |
+
<a id="roadmap"></a>
|
| 56 |
+
|
| 57 |
+
## πΊοΈ Roadmap
|
| 58 |
+
|
| 59 |
+
- [x] [arXiv paper](https://arxiv.org/abs/2608.07462) and [code repository](https://github.com/ASLP-lab/SemBridge)
|
| 60 |
+
- [x] [TTS, SVS, and unified inference code](#inference)
|
| 61 |
+
- [x] [Model checkpoints](https://huggingface.co/ASLP-lab/SemBridge)
|
| 62 |
+
- [ ] Fine-tuning code
|
| 63 |
+
|
| 64 |
+
<a id="installation"></a>
|
| 65 |
+
|
| 66 |
+
## βοΈ Installation
|
| 67 |
|
| 68 |
Use **Python 3.10**.
|
| 69 |
|
|
|
|
| 80 |
python -m pip install -e .
|
| 81 |
```
|
| 82 |
|
| 83 |
+
<a id="inference"></a>
|
| 84 |
+
|
| 85 |
+
## π Inference
|
| 86 |
|
| 87 |
```python
|
| 88 |
from sembridge import SemBridge
|
|
|
|
| 104 |
| **SVS** | Lyrics, MIDI pitches, note durations | BigVGAN-Large |
|
| 105 |
| **Unified** | Ordered speech and singing segments | BigVGAN-Large |
|
| 106 |
|
| 107 |
+
<a id="examples"></a>
|
| 108 |
+
|
| 109 |
+
### π§ Examples
|
| 110 |
|
| 111 |
The repository includes **10 complete examples from our [Demo page](https://tiamojames.github.io/SemBridge_Demo/)**:
|
| 112 |
4 TTS, 4 SVS, and 2 Unified cases, with reference audio and prepared requests.
|
|
|
|
| 131 |
to use local weights. See the **[example catalog](https://github.com/ASLP-lab/SemBridge/blob/main/examples/README.md)** for the
|
| 132 |
complete list, reference recordings, and guidance on using your own songs.
|
| 133 |
|
| 134 |
+
<a id="single-sample-inference"></a>
|
| 135 |
+
|
| 136 |
+
### ποΈ Single-Sample Inference
|
| 137 |
|
| 138 |
```bash
|
| 139 |
# Text-to-speech
|
|
|
|
| 165 |
See the **[inference guide](https://github.com/ASLP-lab/SemBridge/blob/main/docs/INFERENCE.md)** for reference voices, singing and
|
| 166 |
mixed-generation examples, sampling controls, and offline loading.
|
| 167 |
|
| 168 |
+
<a id="community"></a>
|
| 169 |
+
|
| 170 |
+
## π€ Community
|
| 171 |
+
|
| 172 |
+
- **Windows / ComfyUI:** [ComfyUI-SemBridge](https://github.com/hiroki-abe-58/ComfyUI-SemBridge), an unofficial, community-maintained integration.
|
| 173 |
+
|
| 174 |
+
<a id="acknowledgements"></a>
|
| 175 |
+
|
| 176 |
+
## π Acknowledgements
|
| 177 |
|
| 178 |
We thank the authors of the following projects for their open research and resources:
|
| 179 |
|
| 180 |
- **Speech generation:** [VoxCPM](https://github.com/OpenBMB/VoxCPM) and [DiTAR](https://arxiv.org/abs/2502.03930).
|
| 181 |
- **Audio codecs:** [DAC](https://github.com/descriptinc/descript-audio-codec), [SAC](https://github.com/Soul-AILab/SAC), and [BigVGAN](https://github.com/NVIDIA/BigVGAN).
|
| 182 |
|
| 183 |
+
<a id="citation"></a>
|
| 184 |
+
|
| 185 |
+
## π Citation
|
| 186 |
|
| 187 |
If you find SemBridge useful in your research, please cite:
|
| 188 |
|
|
|
|
| 199 |
}
|
| 200 |
```
|
| 201 |
|
| 202 |
+
<a id="license"></a>
|
| 203 |
+
|
| 204 |
+
## π License
|
| 205 |
|
| 206 |
SemBridge code is released under Apache-2.0; upstream portions retain their
|
| 207 |
original licenses. See [LICENSE](https://github.com/ASLP-lab/SemBridge/blob/main/LICENSE) for terms and third-party notices.
|