Inflect-Mini-v3

Inflect Mini v3 has approximately 3431 core parameters, natively outputs 24 kHz audio, and optionally uses a small enhancer to output 48 kHz. It runs on both CPU and CUDA. It comes with six preset English voices, and text can include eight types of sound events such as [laugh], [sigh], [gasp], and [cough]. The Voice Tuner generates reusable voice profiles from authorized reference recordings lasting 5–15 seconds; the author emphasizes that this is reference-based adjustment and does not guarantee exact cloning. The model repository is a complete inference package, and long texts are automatically segmented.

Inflect Mini v3 has approximately 3431 core parameters, natively outputs 24 kHz audio, and optionally uses a small enhancer to output 48 kHz. It runs on both CPU and CUDA. It comes with six preset English voices, and text can include eight types of sound events such as [laugh], [sigh], [gasp], and [cough]. The Voice Tuner generates reusable voice profiles from authorized reference recordings lasting 5–15 seconds; the author emphasizes that this is reference-based adjustment and does not guarantee exact cloning. The model repository is a complete inference package, and long texts are automatically segmented.

Why it matters The model is small and does not rely on cloud interfaces, making it suitable for embedding in local applications; it can also use an authorized recording to create reusable custom voice profiles.

Hugging Face Likes 21 · Hugging Face Trending Score 21 · Hugging Face Downloads 280

License: Commercial Use Allowed · ComfyUI: Not Available · VRAM: Not Specified

Reasons

  • Apache-2.0 license, with original inputs and source descriptions provided for all sample audio
  • An online demo space is available; verify the file manifest after downloading before running

Usage

  • Use a new Python 3.12 environment: hf download owensong/Inflect-Mini-v3 --local-dir inflect-mini-v3, enter the directory and run python verify_release.py, install core/requirements-cpu.txt, then generate audio using python run.py speech --voice cassidy --text "..."

Risks

  • Supports English only; sound events are available only on CPU at 24 kHz, and Voice Tuner currently supports registration only on Linux CPU
  • Voice Tuner requires separately downloaded encoders, each with its own license; you may only use voices for which you have rights

Open

抱脸 text-to-speech 上升榜

Explore the latest AI Radar updates