| --- |
| title: Chatterbox Voiceovers |
| emoji: 📊 |
| colorFrom: pink |
| colorTo: purple |
| sdk: gradio |
| sdk_version: 6.8.0 |
| python_version: '3.10' |
| app_file: app.py |
| pinned: false |
| license: mit |
| short_description: Reel script to voiceover model |
| --- |
| |
| # Chatterbox Voiceovers |
|
|
| This app serves a Gradio interface for turning short Instagram Reel scripts |
| into downloadable WAV voiceovers with Chatterbox Turbo. It is configured to run |
| on CPU by default. |
|
|
| ## Features |
|
|
| - Paste a script up to 2,000 characters. |
| - Select one of three bundled default voice references, or upload/record a |
| custom reference clip for voice cloning. |
| - Insert paralinguistic tags from UI buttons: |
| `[clear throat]`, `[sigh]`, `[shush]`, `[cough]`, `[groan]`, `[sniff]`, |
| `[gasp]`, `[chuckle]`, `[laugh]`. |
| - Automatically pronounce `zapply` as `zap-lee`, case-insensitively. |
| - Preview generated audio and download the WAV output. |
|
|
| ## Bundled Voices |
|
|
| The default voice references are 20-second MP3 excerpts from LibriVox |
| recordings. LibriVox recordings are public domain in the USA; verify copyright |
| status before deploying in jurisdictions with different public-domain rules. |
|
|
| - `assets/voices/elizabeth_klett_female.mp3`: Elizabeth Klett reading |
| `Pride and Prejudice (version 4)`, chapter 2. |
| - `assets/voices/mark_smith_male.mp3`: Mark F. Smith reading |
| `How to Live on Twenty-Four Hours a Day`, chapter 1. |
| - `assets/voices/david_clarke_male.mp3`: David Clarke reading |
| `A Study In Scarlet (Version 6)`, chapter 1. |
|
|
| ## CPU Runtime |
|
|
| The app defaults to `CHATTERBOX_DEVICE=cpu`, so it runs on machines without a |
| CUDA GPU. CPU generation can be slow for longer scripts, so the Gradio queue is |
| limited to one concurrent generation. |
|
|
| To force a different device, set `CHATTERBOX_DEVICE` before launching: |
|
|
| ```bash |
| CHATTERBOX_DEVICE=cuda python app.py |
| ``` |
|
|
