Managing Voices

The Voices tab is your voice library: reference clips you can use for cloning with any model. Save a clip once, and every cloning model can use it under Voice library on the Generate tab.

Bundled voices

OpenVoiceLab ships with four sample voices: Alice, Carter, Frank and Maya. They're copied into your library the first time it's opened, so you can delete or replace them.

Adding a voice

Under Add a voice:

1. Clip: upload an audio file or record one with your microphone.
2. Name: for example narrator.
3. Transcript (optional): what is said in the clip. Leave Transcribe automatically if empty checked, and Whisper fills it in.
4. Click Add voice.

The voice can now be selected under Voice library on the Generate tab.

Why transcripts matter

Some models need the transcript of the reference clip (Qwen3-TTS Base, VoxCPM 1.5 and 0.5B, MOSS-TTSD), and others clone better with it (VoxCPM2, MOSS-TTS). Library voices keep their transcript next to the clip, so it's used automatically.

To fix a transcript, select the voice under Your voices, edit Transcript and click Save transcript. Or click Re-transcribe to run Whisper again. Make sure the transcript matches the audio exactly; mismatches cause skipped or garbled words.

Choosing a good reference clip

Works well:

• 5-20 seconds of one speaker (Echo-TTS can use up to 5 minutes)
• Clean recording with no music, noise or reverb
• Natural, expressive speech in the style you want
• The same language as the text you'll generate

Works badly:

• Background music or other speakers
• Very quiet, clipped or heavily compressed audio
• Whispering or shouting, unless that's the style you want

The model copies whatever it hears, including the recording quality, so a noisy clip gives noisy output.

Where voices are stored

Voices are plain files in the voices/ folder (inside OVL_HOME if you've set it):

voices/
├── narrator.wav
├── narrator.txt      # transcript (optional)
├── podcast_host.mp3
└── podcast_host.txt

Supported formats: WAV, MP3, FLAC, OGG, M4A and Opus. You can copy files into this folder directly; a .txt file with the same name is used as the transcript. To share a voice, send the audio file and its .txt.

Consent

Only clone voices you have permission to use. Don't use cloned voices to impersonate or deceive people.