Voicebox: Seven Engines, One Studio, Your Own Server
A walkthrough of the Voicebox app: speak a long script in any of seven engines, clone a voice with consent, transcribe audio, and mix a voice-over.
Yantra Team2 min read
On this page
Text-to-speech has a dozen good engines now, each best at something. Voicebox puts seven of them in one studio, running on your own server, and adds the two things you always need next: transcription and a mix.
Speak
Paste a script of up to 50,000 characters; it is chunked automatically. Pick a voice: Kokoro presets, Qwen CustomVoice presets, or a voice you cloned. Pick an engine per clip: Kokoro, Qwen CustomVoice, Qwen3-TTS, LuxTTS, Chatterbox Multilingual, Chatterbox Turbo or TADA, across 23 languages. Qwen CustomVoice takes a delivery instruction ("warm, unhurried"), Chatterbox Turbo takes emotion tags, and every clip has a seed so a good take can be reproduced.
For each clip: replay, take again, transcribe, voice-over, send to Canvas, download, copy the URL.
Voices
Filter by kind, engine, language or gender, or search. Try plays a pre-recorded sample. Cloning takes an uploaded file, a microphone recording or a URL, with the exact transcript and a description, and is gated on an explicit consent acknowledgement that you own or have permission to use the voice.
A preset voice speaks only on its own engine. Cloned voices work on Qwen3-TTS, LuxTTS, Chatterbox, Chatterbox Turbo and TADA.
Transcribe
Choose a file, microphone or URL, a Whisper model size from base to large and turbo, and an optional language hint. Speak the result back, or copy it.
Mix
Combine a voice clip with a music bed and/or a video. Set the voice and music weights with ducking, choose whether to replace or keep the video's own audio, and the plan is shown before it runs. Preview, download, or send to Canvas.
From chat
- "Narrate this in a warm British male voice: Welcome back to the show."
- "Make a 20-second lo-fi bed and read my intro over it."
- "Clone my voice from this sample and say welcome to the show."
- "Transcribe this meeting recording."
Setup and limits
The Setup tab holds the server URL, reachability, the model cache and the studio's rules. The server handles one generation at a time, and the first use of an engine downloads its model, which can be several gigabytes. Clip URLs last seven days. Without a connected account the studio runs in demo mode.