Voice Clone

yt-dlp for YouTube/Video Sources

# Download audio only
yt-dlp -x --audio-format mp3 --audio-quality 0 \
  -o "./voice-samples/%(title)s.%(ext)s" \
  "https://youtube.com/watch?v=VIDEO_ID"

# Download specific time range (requires ffmpeg)
yt-dlp -x --audio-format mp3 \
  --postprocessor-args "ffmpeg:-ss 00:01:30 -to 00:03:45" \
  -o "./voice-samples/clip.%(ext)s" \
  "https://youtube.com/watch?v=VIDEO_ID"

Step 2: Prepare Samples

Trim silence, normalize volume, and optionally remove background noise. Requires ffmpeg.

# Prepare all files in a directory
bun run scripts/voice-clone.ts prepare \
  --input-dir ./voice-samples \
  --output-dir ./voice-prepared

# With options
bun run scripts/voice-clone.ts prepare \
  --input-dir ./voice-samples \
  --output-dir ./voice-prepared \
  --trim-silence \
  --normalize \
  --max-duration 60

The script validates ffmpeg is installed and exits with an informative error if not.

Step 3: Clone via IVC

Upload prepared samples to ElevenLabs IVC. The API key must be set in the environment.

# Clone from prepared samples
bun run scripts/voice-clone.ts clone \
  --input-dir ./voice-prepared \
  --name "Movie Announcer" \
  --description "Deep dramatic voice in the style of classic movie trailers" \
  --remove-background-noise

# With labels for organization
bun run scripts/voice-clone.ts clone \
  --input-dir ./voice-prepared \
  --name "Movie Announcer" \
  --description "Deep dramatic voice" \
  --labels '{"accent":"american","age":"middle-aged","gender":"male","use_case":"trailer_narration"}'

The script outputs the voice_id on success. Capture and surface this to the user — it is needed for all subsequent steps.

Step 4: Test the Clone

Generate test speech and output audio files so the user can compare against reference.

# Quick test with default phrases
bun run scripts/voice-clone.ts test \
  --voice-id "VOICE_ID_FROM_STEP_3" \
  --output-dir ./voice-tests

# Test with custom text
bun run scripts/voice-clone.ts test \
  --voice-id "VOICE_ID_FROM_STEP_3" \
  --text "In a world where darkness threatens to consume all hope..." \
  --output-dir ./voice-tests

# Test with specific model
bun run scripts/voice-clone.ts test \
  --voice-id "VOICE_ID_FROM_STEP_3" \
  --model eleven_v3 \
  --output-dir ./voice-tests

Report the output file paths to the user after this step completes.

Step 5: Tune Voice Settings

Adjust stability, similarity boost, and style to dial in the match.

bun run scripts/voice-clone.ts tune \
  --voice-id "VOICE_ID_FROM_STEP_3" \
  --stability 0.3 \
  --similarity-boost 0.8 \
  --style 0.5 \
  --text "In a world where nothing is as it seems..." \
  --output-dir ./voice-tests

When the user does not specify settings, use these defaults: stability 0.5, similarity-boost 0.75, style 0.0. For voice type presets, refer to README.md.

Full Pipeline (One Command)

bun run scripts/voice-clone.ts pipeline \
  --files "./samples/clip1.mp3,./samples/clip2.mp3" \
  --name "Movie Announcer" \
  --description "Deep dramatic voice for movie trailers" \
  --test-text "In a world where heroes are forgotten..." \
  --remove-background-noise \
  --output-dir ./movie-announcer-voice

Runs source → prepare → clone → test in sequence. Output includes the voice_id and paths to test audio files.

Managing Voices

# List all cloned voices
bun run scripts/voice-clone.ts list

# Delete a cloned voice
bun run scripts/voice-clone.ts delete --voice-id "VOICE_ID"

# Get details about a voice
bun run scripts/voice-clone.ts info --voice-id "VOICE_ID"

Error Handling

• If ELEVENLABS_API_KEY is unset, exit immediately with a message directing the user to README.md for setup instructions.
• If ffmpeg is missing, exit with an install prompt (brew install ffmpeg).
• If the IVC API returns a tier error, inform the user that IVC requires Starter tier or above and link to README.md for tier details.
• If voice_id is needed but not yet obtained, prompt the user to complete Step 3 first.