README.md

---
license: mit
language:
- en
base_model:
- ResembleAI/chatterbox
pipeline_tag: text-to-speech
tags:
- gguf-connector
---
## gguf quantized version of chatterbox
- base model from [resembleai](https://huggingface.co/ResembleAI)
- text-to-speech synthesis

### connector c3 (*multilingual update)
```
ggc c3
```

*note: for c3, you might need to downgrade `transformers` to 4.46.3 (c2 below compatible with the latest version); upgrade `chichat` to 0.0.4, and opt `t3_23lang` clip instead of `t3_cfg`

![screenshot](https://raw.githubusercontent.com/calcuis/text-to-speech-synthesis-lite/master/demo2.png)

### **run it with gguf-connector**
```
ggc c2
```

![screenshot](https://raw.githubusercontent.com/calcuis/text-to-speech-synthesis-lite/master/demo.png)

| Prompt | Audio Sample |
|--------|---------------|
|`Hey Connector, why your appearance looks so stupid?`<br/>`Oh, really? maybe I ate too much smart beans.`<br/>`Wow. Amazing.`<br/>`Let's go to get some more smart beans and you will become stupid as well.`<br/> | 🎧 **audio-sample-1**<br><audio controls src="https://huggingface.co/calcuis/chatterbox-gguf/resolve/main/samples%5Caudio1.wav"></audio> |
|`Now let's make my mum's favourite. So three mars bars into the pan. Then we add the tuna and just stir for a bit, just let the chocolate and fish infuse. `<br/>`A sprinkle of olive oil and some tomato ketchup. Now smell that. Oh boy this is going to be incredible.`<br/> | 🎧 **audio-sample-2**<br><audio controls src="https://huggingface.co/calcuis/chatterbox-gguf/resolve/main/samples%5Caudio2.wav"></audio> |

### **review/reference**
- simply execute the command (`ggc c2`) above in console/terminal
- opt a `vae`, a `clip(encoder)` and a `model` file in the current directory to interact with (see example below)

>
>GGUF file(s) available. Select which one for **ve**:
>
>1. s3gen-bf16.gguf
>2. s3gen-f16.gguf
>3. s3gen-f32.gguf
>4. t3_cfg-q2_k.gguf
>5. t3_cfg-q4_k_m.gguf
>6. t3_cfg-q6_k.gguf
>7. ve_fp32-f16.gguf (recommended)
>8. ve_fp32-f32.gguf
>
>Enter your choice (1 to 8): 7
>
>ve file: ve_fp32-f16.gguf is selected!
>
>GGUF file(s) available. Select which one for **t3**:
>
>1. s3gen-bf16.gguf
>2. s3gen-f16.gguf
>3. s3gen-f32.gguf
>4. t3_cfg-q2_k.gguf
>5. t3_cfg-q4_k_m.gguf (recommended)
>6. t3_cfg-q6_k.gguf
>7. ve_fp32-f16.gguf
>8. ve_fp32-f32.gguf
>
>Enter your choice (1 to 8): 5
>
>t3 file: t3_cfg-q4_k_m.gguf is selected!
>
>GGUF file(s) available. Select which one for **s3gen**:
>
>1. s3gen-bf16.gguf (recommended)
>2. s3gen-f16.gguf (for non-cuda user)
>3. s3gen-f32.gguf
>4. t3_cfg-q2_k.gguf
>5. t3_cfg-q4_k_m.gguf
>6. t3_cfg-q6_k.gguf
>7. ve_fp32-f16.gguf
>8. ve_fp32-f32.gguf
>
>Enter your choice (1 to 8): _
>

- note: for the latest update, only tokenizer will be pulled to cache automatically during the first launch; you need to prepare the **model**, **encoder** and **vae** files yourself, working like [vision](https://huggingface.co/calcuis/llava-gguf) connector right away; mix and match, more flexible
- run it entirely offline; i.e., from local URL:  http://127.0.0.1:7860 with lazy webui
- gguf-connector ([pypi](https://pypi.org/project/gguf-connector))