| # Sample audio files |
|
|
| Place six WAV files here so the `/api/samples` endpoint can serve them and the |
| Demo page's "Sample Library" can play them in-browser: |
|
|
| | File | Label | Suggested source | |
| |------|-------|------------------| |
| | `real_news_excerpt.wav` | REAL | LibriSpeech / VCTK clean recording | |
| | `real_conversation.wav` | REAL | Common Voice spontaneous clip | |
| | `real_lecture.wav` | REAL | TED-LIUM lecture excerpt | |
| | `fake_tts_commercial.wav` | FAKE | Eleven Labs / Microsoft TTS export | |
| | `fake_voice_clone.wav` | FAKE | Few-shot voice conversion (RVC, So-VITS) | |
| | `fake_neural_tts.wav` | FAKE | Coqui TTS / xTTS-v2 output | |
|
|
| ## Format requirements |
| - 16 kHz, mono, PCM 16-bit (the backend resamples anything else, but this is |
| the canonical format). |
| - 3–10 s each is plenty. |
| - Keep file size under 1 MB so the demo loads quickly. |
|
|
| ## How to populate |
|
|
| 1. Drop your WAVs in this folder using exactly the filenames above. |
| 2. Restart the backend (`docker compose restart backend` or `uvicorn`). |
| 3. Visit `/api/samples` — your clips will appear with the correct labels. |
|
|
| If you don't have ready-made clips, the easiest pipeline is: |
| - Record yourself for the REAL files via the in-app microphone, then export |
| the saved clips and rename them. |
| - Generate the FAKE files using any open-source TTS (e.g. [Coqui xTTS-v2](https://github.com/coqui-ai/TTS), |
| [Tortoise TTS](https://github.com/neonbjb/tortoise-tts), or a hosted service |
| such as Eleven Labs). |
|
|
