voice clone
Clone a voice from an audio clip. Returns a voice ID for use with
'egaki speech --voice <id>'. Supports Cartesia (default) and ElevenLabs.
Best practices for high-quality clones:
- Isolate vocals first: 'egaki demucs recording.mp3 --stems vocals' removes background music, noise, and other speakers.
- Find a clean snippet: 'egaki transcribe recording-vocals.mp3' to get word timestamps. Pick a 5-10s segment with a complete phrase, clear speech, no hesitations or crosstalk.
- Trim to speech boundaries: ffmpeg -i recording-vocals.mp3 -ss 12.5 -to 22.0 -c copy clip.mp3 No silence padding at start or end.
- Match energy to intent: the clone mimics the tone and pacing of the source clip. Use an energetic clip for energetic output.
- Speak in the target language. Use --language for Cartesia clones.
Cartesia: up to 10s of audio, instant, free. Good for short clean clips.
ElevenLabs: 1-3 min recommended, has --remove-background-noise option.
Usage
1egaki voice clone [audio]
Arguments
| Argument | Required | Description |
[audio] | No | audio |
Options
| Option | Default | Description |
--name [name] | - | Name for the cloned voice (required) |
-p, --provider [provider] | cartesia | Voice cloning provider: cartesia or elevenlabs |
--language [lang] | - | Cartesia only: ISO 639-1 language code (default: en). E.g. en, es, fr, de, ja |
--description [text] | - | Optional description for the voice |
--base-voice-id [id] | - | Cartesia: optional base voice ID to derive from |
--remove-background-noise | - | ElevenLabs: apply AI noise removal to the clip before cloning |
--stdin | - | Read audio from stdin instead of a file path |
--json | - | Output result as JSON to stdout |
Global Options
| Option | Default | Description |
-h, --help | - | Display this message |
-v, --version | - | Display version number |
Examples
1# Clone a voice from a recording
1egaki voice clone recording.wav --name "My Voice"
1# Clone with ElevenLabs and noise removal
1egaki voice clone vocals.mp3 --name "Narrator" --provider elevenlabs --remove-background-noise
1# Full pipeline: separate → trim → clone
1egaki demucs interview.mp3 --stems vocals
1ffmpeg -i interview-vocals.mp3 -ss 5.0 -to 15.0 -c copy clip.mp3
1egaki voice clone clip.mp3 --name "Speaker"
1# Use the cloned voice
1egaki speech "Hello world" --voice <voice-id>