AIAny
AI Audio2025
Icon for item

VieNeu-TTS

Synthesizes Vietnamese and English–Vietnamese speech locally, with 3–8-second voice cloning, preset speakers, and streaming. The open v3 Turbo runs on CPU or GPU; v4 is proprietary and API-only.

Introduction

For languages with fewer speech tools, practical access matters as much as synthesis quality: a model that needs a GPU server can be out of reach for local apps and smaller teams. This project’s notable bet is to make Vietnamese speech generation usable on-device, while also supporting English–Vietnamese code-switching and short-clip voice cloning.

What Sets It Apart
  • Local CPU and GPU inference: The open v3 Turbo supports an ONNX-based CPU path as well as GPU inference, making deployment possible without a dedicated speech API. Its streaming mode is designed for low-latency applications.
  • Short-reference cloning: A few seconds of reference audio can provide a speaker voice for generated speech. This lowers the amount of recording needed, but also means teams should handle permission and misuse risks carefully.
  • Vietnamese-first, bilingual design: Training on more than 10,000 hours of English–Vietnamese speech aims to support both Vietnamese and code-switched text, rather than treating Vietnamese as an afterthought.
  • Open and proprietary versions differ: v3 Turbo is the latest open-source release in the repository. The newer v4 is proprietary and available through the VieNeu service, so it is not a self-hostable upgrade.
Who It’s For

A good fit for developers building Vietnamese narration, voice interfaces, or local speech tools who want to compare CPU deployment with GPU throughput and can evaluate voice quality on their own data. Look elsewhere if you need a fully open, self-hosted version of the newest model, or if your target language is not Vietnamese or English–Vietnamese; the preview Nano variant is faster on weak CPUs but trades away audio quality, especially for English.

Information

  • Websitegithub.com
  • AuthorsPhạm Nguyễn Ngọc Bảo
  • Published date2025/10/28

More Items

Hugging Face
AI Model2025

Labels who speaks when in an audio recording, with improved speaker assignment and counting. It runs locally, supports offline use, and offers an exclusive diarization output that is easier to align with transcription timestamps.

Hugging Face
AI Model2026

An experimental Qwen3.8-27B variant edited in weight space to use fewer reasoning tokens; reported evaluations show higher accuracy on GPQA Diamond and LiveCodeBench v6.

Hugging Face
AI Model2026

Listens and responds during live speech, interpreting cues such as hesitation, emphasis and sarcasm. Its replies can include expressive nonverbal sounds; Tag, Flash and E2E builds trade expressiveness, throughput and first-audio latency.