Metadata-Version: 2.1
Name: haaztest-kitten-inference-tts2-cu12
Version: 0.2.0
Summary: Kitten TTS 2 native engine for NVIDIA GPUs (CUDA 12.8): library and command-line tool
Home-page: https://github.com/KittenML/kitten-inference
Author: KittenML
License: Apache-2.0 for KittenML's code. The library also contains third-party code under other licenses (LGPL-2.1-or-later, BSD-3-Clause, MIT, NVIDIA CUDA EULA and more), listed in NOTICE; the whole is not Apache-2.0-only
Classifier: Environment :: GPU :: NVIDIA CUDA :: 12
Classifier: Programming Language :: Python :: 3
Classifier: Operating System :: POSIX :: Linux
Classifier: Topic :: Multimedia :: Sound/Audio :: Speech
Requires-Python: >=3.9
Description-Content-Type: text/markdown

# kitten-inference-tts2-cu12

The Kitten TTS 2 native engine for NVIDIA GPUs, with its command-line tool. On an RTX PRO 6000 with the current
model it streams its first audio after about **11 ms**, and complete files take 5 to 7 times less time than with
Kitten TTS 2 on vLLM or llama.cpp (one request at a time).

Install it with the Python API, which uses it:

```sh
python3 -m venv .venv && . .venv/bin/activate   # Windows: py -m venv .venv, then .venv\Scripts\activate
python -m pip install -U pip                    # pip older than 23.3 skips the "[tts-0.8]" extra with only a warning
pip install "kitten-inference[tts-2]"
```

| NVIDIA GPU | Status |
|---|---|
| Compute capability 8.0 or newer (GeForce RTX 30 series and newer) | supported |
| GeForce GTX 10, GTX 16 and RTX 20 series (compute capability 6.1 and 7.5) | supported, not optimal: no bf16 tensor cores, so part of the math runs in software. Real-time factor 0.22 to 0.76 with `--stream false`, against 0.03 to 0.17 for RTX 30 series and newer. The audio is close to, not bit for bit the same as, a newer GPU's |

It needs NVIDIA driver R570 or newer; the CUDA 12.8 runtime is built in. Voice cloning reads the reference audio with
the `ffmpeg` program (`sudo apt-get install -y ffmpeg`, or on Windows `winget install --id Gyan.FFmpeg -e`).
