nghimestudio/vieneu-tts

VieNeu-TTS v3 Turbo — Vietnamese TTS with 25 preset voices, instant cloning, emotion cues, 48 kHz audio.

Public
52 runs

Run time and cost

This model runs on CPU hardware. We don't yet have enough runs of this model to provide performance information.

Readme

VieNeu-TTS v3 Turbo

Vietnamese (and bilingual) text-to-speech with instant voice cloning, powered by VieNeu-TTS.

48 kHz high-fidelity speech, 25 built-in preset voices, optional zero-shot cloning from a short reference clip, and experimental emotion tags like [cười], [thở dài], [hắng giọng].

Examples

Preset voice

Input - text: Xin chào! Đây là giọng đọc VieNeu-TTS v3 Turbo trên Replicate. - voice: Hải Đăng

Output: 48 kHz WAV in the Hải Đăng preset voice.

With emotion cue

Input - text: [cười] Trời ơi, cái giọng nó tự nhiên quá. Giờ thì tha hồ mà quẩy content nhé! - voice: Mai Anh

Output: Natural Vietnamese speech with a chuckle cue.

Instant voice cloning

Input - text: Đây là giọng được nhân bản tức thì từ một đoạn thu âm ngắn. - ref_audio: a clean 3–8s reference WAV of the target speaker - denoise: true

Output: Cloned-speaker WAV at 48 kHz.

Inputs

Name Type Default Description
text string required Vietnamese / bilingual text (supports [cười], [thở dài], [hắng giọng])
voice choice Hải Đăng One of 25 built-in preset voices
ref_audio file optional 3–8s reference clip for cloning (overrides voice)
denoise bool true Denoise / trim the reference before cloning
temperature float 0.8 Sampling temperature
top_k int 25 Top-k sampling
top_p float 0.95 Nucleus sampling
repetition_penalty float 1.2 Repetition penalty
max_chars int 256 Max characters per synthesis chunk

Tips

  • Start with editors’ picks: Hải Đăng, Mai Anh, Trúc Ly, Adam bựa, Ngọc Lý / Ngọc Huyền.
  • Keep clone references short (3–8s), clean, and mostly continuous speech.
  • Long text is chunked automatically via max_chars.
Model created
Model updated