kitten-tts-2-mini

Kitten TTS 2 Mini has approximately 8200 ten-thousand parameters, supports 16 languages including English, German, Spanish, French, Japanese, and Korean, offers a total of 118 voices, and outputs audio at 24 kHz. By default, it runs int8 weights using ONNX Runtime (82.7 MB, CPU only, no PyTorch required), but float32 PyTorch weights are also provided for GPU execution. The English voices are character-based personas (such as a deep-voiced historian or an enthusiastic sports commentator); switching the language parameter allows the same voice to speak five other languages.

Kitten TTS 2 Mini has approximately 8200 ten-thousand parameters, supports 16 languages including English, German, Spanish, French, Japanese, and Korean, offers a total of 118 voices, and outputs audio at 24 kHz. By default, it runs int8 weights using ONNX Runtime (82.7 MB, CPU only, no PyTorch required), but float32 PyTorch weights are also provided for GPU execution. The English voices are character-based personas (such as a deep-voiced historian or an enthusiastic sports commentator); switching the language parameter allows the same voice to speak five other languages.

Why it matters It works without installing PyTorch, making it suitable for embedding in local applications; under the same name, English voices can also speak German, Spanish, Italian, Portuguese, and Russian.

Hugging Face Likes 17 · Hugging Face Trending Score 17 · Hugging Face Downloads 1271

License: Commercial use allowed · ComfyUI: Not available · VRAM: ≈1 GB (estimated)

Reasons

  • Apache-2.0 license applies to the model, voice packs, and included pronunciation files
  • After running pip install kittenml, you can generate audio with just a few lines of code

Usage

  • Run pip install kittenml, then use KittenTTS("KittenML/kitten-tts-2-mini").generate("Hello there", voice="Eleanor"), and save the output at 24000 sample rate using soundfile; add backend="torch", device="cuda" to switch to GPU

Risks

  • Does not support Chinese; Hindi, Japanese, Korean, Turkish, etc., have only 2–5 voices
  • The model card does not provide audio quality benchmarks or sample audio

Open

抱脸 text-to-speech 上升榜

Explore the latest AI Radar updates