# yt-builder

Build a Qwen-TTS voice-clone profile from a YouTube video.

## Setup

```bash
pip install -r requirements.txt
```

## Usage

```bash
python build_profile.py <youtube_url> <profile_name> [options]
```

### Options

| Flag | Default | Description |
|------|---------|-------------|
| `--max-duration` | `30` | Max audio duration (seconds) |
| `--start` | — | Start offset in audio (seconds) |
| `--end` | — | End offset in audio (seconds) |
| `--profiles-dir` | `../profiles` | Output profiles directory |
| `--whisper-model` | `large-v3` | Whisper model size |
| `--device` | `cuda` | `cuda` or `cpu` |
| `--language` | auto | Force source language (e.g. `en`, `ar`) |
| `--denoise` | off | Remove background music/noise before processing |
| `--denoise-method` | `demucs` | `demucs` (vocal isolation from music) or `noisereduce` (spectral gating for ambient noise) |
| `-v` | — | Verbose logging |

### Example

```bash
python build_profile.py \
    "https://www.youtube.com/watch?v=ABC123" my-speaker \
    --max-duration 25 --start 5 --end 35
```

Output lands in `collections/profiles/my-speaker/` with `voice.wav` and `script.txt`, ready for:

```bash
mazinger-dubber dub <video> --clone-profile my-speaker
```