VieNeu-TTS v3 Turbo CPP

This repository contains the model asset package for running VieNeu-TTS v3 Turbo with the VieNeu-TTS.cpp C++ runtime.

It is not a standalone Transformers checkpoint and it is not a single-file GGUF model. The package combines a GGUF semantic backbone, native runtime weights, tokenizer/config files, voice presets, and MOSS codec ONNX assets expected by the C++ vieneu-v3-native pipeline.

Files

config.json
tokenizer.json
voices_v3_turbo.json
backbone.gguf
vieneu_v3_heads.npz
acoustic/vieneu_acoustic_weights.npz
codec/moss_audio_tokenizer_decode_full.onnx
codec/moss_audio_tokenizer_decode_shared.data
codec/moss_audio_tokenizer_encode.onnx
codec/moss_audio_tokenizer_encode.data

Download

Use the Hugging Face CLI or Python Hub client to download the full folder:

hf download lastudio-community/VieNeu-TTS-v3-Turbo-CPP --local-dir .models/vieneu-v3-turbo-native

Usage With VieNeu-TTS.cpp

Pass the downloaded folder as the model directory and use the bundled codec folder:

.\build\Release\vieneu-tts-cli.exe `
  --profile vieneu-v3-native `
  --model-dir .models\vieneu-v3-turbo-native `
  --codec-dir .models\vieneu-v3-turbo-native\codec `
  --text "Xin chao, day la VieNeu-TTS v3 Turbo chay bang C++." `
  --output outputs\vieneu-v3-turbo-cpp.wav

The helper benchmark scripts in VieNeu-TTS.cpp also default to this folder layout.

License

This package is distributed under Apache License 2.0, matching the upstream model license.

When you reuse, redistribute, or convert these assets, please keep the license notice and attribution intact for both the original project and this Hugging Face package:

• pnnbao97/VieNeu-TTS
• lastudio-community/VieNeu-TTS-v3-Turbo-CPP

If you bundle additional third-party assets, their own licenses still apply as well.

Credits

Thanks to the original VieNeu-TTS author and maintainers for the model architecture, training, and upstream assets that this C++ package is built around.

Notes

• backbone.gguf is only one part of the runtime package.
• The .npz files are stored without compression because the current C++ loader expects stored NPZ entries.
• The codec/ directory is required for reference audio encoding and waveform decoding.
• This package is intended for the C++ runtime path, not direct loading with transformers.AutoModel.