AI · AI model
Pocket TTS
Lightweight, CPU-friendly text-to-speech model with fast voice cloning.
Pocket TTS is an open-source 100M-parameter speech synthesis model developed by Kyutai Labs. It generates high-quality speech faster than real-time on standard laptop CPUs without requiring a GPU. It bridges the gap between massive server-hosted TTS models and tiny fixed-voice models by supporting rapid voice cloning from short audio references.
What it's for
- Synthesize speech locally on laptops and edge hardware without GPU infrastructure
- Clone speaker voices using a short 5-second audio reference clip
- Integrate fast, streaming text-to-speech into local CLI tools or Python applications
Features
- Runs faster than real-time on standard CPUs (demonstrated over 3x real-time on Apple Silicon)
- Continuous Audio Language Model (CALM) architecture predicting continuous latents
- Voice cloning capturing tone, acoustic conditions, and accents from short samples
- One-step sampling via Lagrangian Self-Distillation (LSD) and drifting heads
- Simple CLI interface executable directly via uvx
- Multilingual support with checkpoints for English, French, Italian, German, Spanish, and more
- Comes with predefined voice embeddings and evaluation scripts
Pricing
Open source. Free and open source under the MIT license.
Details
- Type
- AI model
- Category
- AI
- Topic
- Speech Synthesis
- Open source
- Yes
- Platforms
- Command line, Code library, macOS, Linux, Windows
- For
- Developers, AI researchers, and hobbyists looking for fast, local text-to-speech with voice cloning.
- Alternative to
- Kokoro, F5-TTS, Chatterbox Turbo, ElevenLabs
- Tags
- tts, text to speech, voice cloning, audio, cpu inference, open source
Why it was saved
- @kyutai_labs: Bridges the gap between huge GPU-bound LLM TTS models and small models like Kokoro (82M) that lack flexible voice cloning. True voice cloning from just 5 seconds of audio: captures voice color, emotion/accent, and acoustic conditions (reverb, mic quality). Runs via a simple CLI (`uvx pocket-tts generate --text ... --voice sample.wav`).
Compare
Related
Open-source deep learning toolkit for text-to-speech synthesis and voice cloning.
AI text-to-speech, instant voice cloning, and emotional speech control
AI voice generator offering realistic narration, voice cloning, and speech-to-speech tools.
Open-source zero-shot text-to-speech and voice cloning model
AI text-to-speech platform with multilingual voices and voice cloning.
AI voice generator, voice cloning, and conversational agent platform
Updated