duarteocarmo/voice/dataset-builder - Storage Bucket

Dataset Builder

Minimal local web app to record prompted TTS fine-tuning data.

Run

bun run dev

Open http://localhost:3000.

Use

1. Put one prompt per line in sentences.txt, or upload a .txt in the UI.
2. Pick an audio source if needed.
3. Click Record, speak the sentence, then Stop.
4. Listen back and click Save or Save + Next.

Shortcuts:

• R or Space: record/stop
• L: listen
• S: save
• Enter: save + next

Outputs:

• WAV files: recordings/utt_000001.wav, ...
• Qwen-style metadata: metadata.jsonl

Each metadata line looks like:

{"audio":"./recordings/utt_000001.wav","text":"She said she would be here by noon.","ref_audio":"./reference/ref.wav"}

Put your shared reference clip at reference/ref.wav, or edit metadata.jsonl afterwards.