Voice engines
| Engine | Characteristics |
|---|---|
| Gemini TTS (default) | Expressive reading with natural emphasis. The channel's standard voice: Charon (male) |
| VieNeu | Runs locally, 10 Vietnamese voices, voice cloning from a 3-5 second sample |
| Edge TTS | Quick free fallback |
Reading speed is 1.0-1.2×. If you don't pass voice, the app still uses the Charon voice.
Generate voice content from a link
- Single-voice narration (
/api/v1/tts/summary): paste a product link → a short, medium or long summary script → read aloud. - Two-host podcast (
/api/v1/tts/podcast): natural NotebookLM-style dialogue with filler words, short interjections and back-and-forth reactions.
Speech recognition (transcribe)
faster-whisper runs locally for free: POST /api/v1/transcribe returns text with word-level timestamps; /api/v1/transcribe/srt exports .srt subtitles. This is also the engine behind the "show it as it's said" effect in videos.
Review the script before rendering
When you need to approve the wording, export the transcript and read it first - this avoids rendering a video (which spends voice quota) just to check the lines.