Clone custom voices for VoiceMode TTS using local mlx-audio and reference audio clips.
npx skills add https://github.com/mbailey/voicemode --skill impressions
Please help me install this Agent Skill. Skill: impressions Source: https://github.com/mbailey/voicemode/tree/main/.claude/skills/impressions Command: npx skills add https://github.com/mbailey/voicemode --skill impressions
VoiceMode's built-in Kokoro and OpenAI voices are fixed; this Skill lets you make VoiceMode speak in any voice by cloning it from a short reference audio clip using local Qwen3-TTS on mlx-audio.
voicemode clone add, then use it via voicemode converse --voice <name> or the MCP converse tool.voice.md profile.VOICEMODE_REMOTE_VOICES_DIR.voice="fleabag" to converse.Ask the assistant to install the mlx-audio service and clone a voice from a short reference clip so VoiceMode can speak in that voice during conversations.
High-intent search queries and answers about installing and using this skill.
Install the mlx-audio service with `voicemode service install mlx-audio`, then run `voicemode clone add <name> <clip.wav>` with a 3-9 second reference clip. Use the voice by passing `voice="<name>"` to the converse tool.
Use 5-9 seconds of clean, conversational speech from a single speaker with no music, cross-talk, or background noise. Always pair the clip with an accurate transcript, since a missing transcript causes stammered synthesis.
No, the mlx-audio service is Apple Silicon only and refuses to install on Intel Macs, Linux, or Windows. On those platforms, use cloud TTS providers like OpenAI or point VoiceMode at a remote Apple Silicon mlx-audio server.
Stammering happens when the reference clip lacks a transcript, forcing the model to transcribe the audio itself and mis-hear it. Fix it by adding a corrected transcript in `voice.md`, a `<clip>.txt` file, or the `ref_text` parameter in converse.
The voice profile is not being recognized. Check that `~/.voicemode/voices/<name>/default.wav` exists and that the name does not collide with a built-in Kokoro voice like `af_sky`, which shadows the custom profile.
Set `VOICEMODE_MLX_AUDIO_BASE_URL` to the remote server's URL and `VOICEMODE_REMOTE_VOICES_DIR` to the voices path as seen by that server. VoiceMode rewrites reference audio paths so the remote host can resolve them on its own filesystem.
akrindevTrim video segments to exact start and end times using FFmpeg.
psd401Generate spoken audio from text using the Eleven Labs API.
damionrashfordAutomate 2-pass EBU R128 loudness normalization for audio directories with ffmpeg-normalize.
bolander72Converts text to speech using local models on Apple Silicon.