Models
Voice models Boubou runs on your own machine. Pick one, copy its command and paste it into a terminal, or into the Developer entry of the boubou menu.
12 models
Integrated
One model hears, thinks and speaks. Served on Mac and Linux.
Liquid AI’s audio model. The default, and the smallest way to start talking.
English voice chat, plus text answers about images and short videos.
A smaller English voice model. Interrupting it takes longer, because Boubou restarts its server.
Pipeline
Hugging Face speech-to-speech: voice activity, speech to text, language model, text to speech. Most also run on Linux with an NVIDIA GPU.
Qwen3-TTS CustomVoice with 9 preset speakers. Whisper versions hear and answer in your language.
Kyutai’s Pocket TTS on the CPU, with a stock config for each of seven languages and 27 preset voices.
Phonon S, a tiny English voice from Gradium that runs on the CPU, with 37 English catalog voices.
Kokoro-82M: 54 voices in 8 languages. Boubou speaks American English as af_heart.
KittenTTS ONNX checkpoints from 15M to 80M parameters, with eight voices on the CPU.
Supertonic 3 on ONNX Runtime’s CPU provider, with 10 voice styles and 31 languages.
ChatTTS speaks English and Chinese, always with the same speaker.
OmniVoice 0.6B, a massively multilingual voice that can design or clone voices.
Meta’s MMS English voice, a VITS model on the CPU.
No model fits these filters
Try another language, more memory, a larger download or another system.