🎙️ Voice Cloner Studio

Natural voice cloning in 23 languages, including Hindi. You get a browser GUI, a one-line command and a simple Python API. It is powered by Chatterbox Multilingual from Resemble AI, an open model (MIT licence) whose naturalness is in the same league as commercial services.

pip install voice-cloner-studio
voice-cloner-studio            # opens the GUI at http://127.0.0.1:7860

Python, in 4 lines

from voice_cloner_studio import VoiceCloner

vc = VoiceCloner()
vc.clone("my_voice", "my_voice.wav")                                  # once - saved for next time
vc.speak("Hello everyone! Welcome to my channel.", voice="my_voice")  # -> dict with "path", "similarity"
vc.speak("नमस्ते दोस्तों!", voice="my_voice", language="Hindi", style="Expressive / energetic")

Command line

voice-clone --ref my_voice.wav --voice me --text "Hello there!" --out hello.wav
voice-clone --voice me --language Hindi --text "नमस्ते, आप कैसे हैं?"
voice-clone --voice me --text-file story.txt --style "Calm narrator" --takes 3

Features

• Clone a voice from a clip or a microphone recording. The app cleans the clip, normalizes loudness and shows a quality report.
• Choose a style: Natural, Expressive, Calm narrator, Fast speaker, or Accent fix for speaking another language. You can also fine-tune emotion, pace and variation.
• Best of N takes: the app keeps the take that sounds most like the voice. A built-in speaker-verification model scores how close each result is.
• Built-in default voice: you can generate speech without cloning anything.
• Long texts are split into sentences automatically. A blank line adds a paragraph pause.
• Voice library: clone once and reuse forever.
• Clear errors: the GUI shows the real reason, and full details go to error.log.

Voices and outputs are saved in VoiceClonerStudio in your home folder. Change this with --data-dir. The model (~2 GB) downloads once, on first use.

Best results

An NVIDIA GPU makes generation real-time. On a laptop CPU, expect about 3-10 s of waiting per second of speech.

Responsible use

Clone only your own voice or voices you have permission to use. Chatterbox adds an inaudible watermark (Perth) to generated audio, so it can be identified as AI-made.

Licence

The code is under the MIT licence. Chatterbox is MIT, and the bundled Resemblyzer speaker encoder is Apache-2.0.