pnnbao97/VieNeu-TTS

Vietnamese TTS with instant voice cloning • On-device • Real-time CPU inference • 48kHz audio quality • Chuyển văn bản thành giọng nói tiếng Việt • Text to speech tiếng Việt • TTS tiếng Việt

What it solves

VieNeu-TTS provides a high-fidelity, low-latency text-to-speech (TTS) system specifically optimized for the Vietnamese language. It addresses the need for natural-sounding Vietnamese speech synthesis that can run efficiently on consumer hardware, including CPUs and mid-range GPUs, while supporting instant voice cloning and real-time streaming.

How it works

The project uses a custom architecture (v3 Turbo) trained on over 10,000 hours of bilingual data. It employs a flow-matching model and a specialized audio tokenizer (MOSS-Audio-Tokenizer-Nano) and phonemizer (sea-g2p). For deployment, it offers multiple backends: a PyTorch engine for GPU acceleration with CUDA graphs for maximum throughput, and a torch-free ONNX Runtime path for CPU execution. It also includes a v3 Nano version for edge devices that trades some audio quality for significantly higher speed.

Who it’s for

  • Developers building Vietnamese AI agents, voice assistants, or accessibility tools.
  • Content Creators needing natural Vietnamese voiceovers for videos, podcasts, or audiobooks.
  • Edge Computing Engineers deploying TTS on low-power ARM boards or old laptops.

Highlights

  • Extreme Performance: Real-time streaming with first audio in ~115ms and a Real-Time Factor (RTF) as low as 0.01–0.02 on a consumer GPU.
  • Instant Voice Cloning: Ability to clone a voice from a 3–8 second reference clip with automatic denoising.
  • OpenAI-Compatible API: A drop-in replacement for the OpenAI TTS endpoint, making it compatible with SDKs like Pipecat and LiveKit.
  • High Fidelity: Supports 48 kHz audio output and includes 25 built-in preset voices across Northern, Central, and Southern Vietnamese regions.
  • Experimental Emotion Cues: Supports inline tags like [cười] (chuckle) and [thở dài] (sigh) to add non-verbal expressions.

Related

  • Project
    tronghieuit/v-tts

    A lightweight Vietnamese text-to-speech system featuring multi-speaker synthesis and zero-shot voice cloning that runs efficiently on CPU.

  • Dispatch
    Qwen3-TTS-VD-Flash and Qwen3-TTS-VC-Flash release: controllable voice design and rapid multilingual voice cloning

    Qwen released Qwen3‑TTS‑VD‑Flash for natural‑language voice design and Qwen3‑TTS‑VC‑Flash for 3‑second multilingual voice cloning, both outperforming leading TTS systems on controllability and accuracy.

  • Project
    kyutai-labs/pocket-tts

    Pocket TTS is a 100 M‑parameter, CPU‑only text‑to‑speech library that streams audio with ~200 ms latency, supports multiple languages and voice cloning, and offers a Python API, CLI, and local HTTP server. It can be installed via pip (or uv) and runs faster than real‑time on modest hardware, with community‑trained models and several alternative runtimes for browsers and embedded devices.

  • Project
    OpenMOSS/MOSS-TTS-Nano

    A multilingual, tiny speech generation model with 0.1B parameters designed for real-time, CPU-based inference and high-fidelity voice cloning.

  • Project
    sipeter/CloneTTS

    An offline Android TTS engine that enables local voice cloning from short audio samples for private, on-device text-to-speech synthesis.