Irodori-TTS-v4.1-Small-Quantized

Irodori-TTS-v4.1-Small-Quantized

Irodori-TTS-v4.1-Small-Quantized provides post-training torchao quantized variants of Aratako/Irodori-TTS-v4.1-Small, a Japanese Text-to-Speech model supporting voice cloning, text-based Voice Design, style-controlled voice cloning, long reference audio, and emoji-based control.

Quantization Variants

These checkpoints quantize compatible attention and MLP Linear layers in the shared text encoder, speaker encoder, and diffusion Transformer. Projectors, AdaLN, duration prediction, and other unsupported layers remain in BF16.

Runtime validation was performed with NVIDIA CUDA. CPU, ROCm, and Intel XPU execution depends on the available PyTorch and torchao kernels and has not been validated for this release. Use --model-precision bf16 for inference.

Usage

For installation instructions and complete CLI, Gradio, LoRA, and long-reference usage, see the Irodori-TTS GitHub repository.

Install the CUDA 12.8 environment:

git clone https://github.com/Aratako/Irodori-TTS.git
cd Irodori-TTS
uv sync --extra cu128

Select a quantization variant by appending its subfolder name to the Hugging Face repo id:

uv run --no-sync python infer.py \
  --hf-checkpoint Aratako/Irodori-TTS-v4.1-Small-Quantized/int8-weight-only \
  --model-precision bf16 \
  --text "こんにちは、私はAIです。これは音声合成のテストです。" \
  --no-ref \
  --output-wav outputs/sample.wav

Replace int8-weight-only with any variant listed in the table above. Only the selected model variant and the shared tokenizer assets are downloaded.

License & Ethical Restrictions

License

This model is released under MIT.

Ethical Restrictions

In addition to the license terms, the following ethical restrictions apply:

1. No Impersonation: Do not use this model to clone or impersonate the voice of any individual (e.g., voice actors, celebrities, public figures) without their explicit consent.
2. No Misinformation: Do not use this model to generate deepfakes or synthetic speech intended to mislead others or spread misinformation.
3. Voice Generation Disclaimer: When generating speech purely from text or captions without using reference audio, it is possible that the generated voice may coincidentally resemble that of a real person. This is strictly a probabilistic artifact within the latent space. The model was not trained with the intent of reproducing specific individuals.
4. Liability Disclaimer: The developers assume no liability for any misuse of this model. Users are solely responsible for ensuring their use of the generated content complies with applicable laws and regulations in their jurisdiction.