SDK Overview

The VieNeu-TTS SDK provides a Python interface for text-to-speech synthesis.

Architecture

Vieneu(mode=...)          # Factory function
  ├── "standard" → VieNeuTTS        (GGUF / PyTorch)
  ├── "fast"     → FastVieNeuTTS    (LMDeploy GPU)
  ├── "remote"   → RemoteVieNeuTTS  (API client)
  └── "xpu"      → XPUVieNeuTTS    (Intel Arc GPU)

Choosing a Mode

Mode	Backend	Best For
standard	GGUF (CPU) or PyTorch	Default, works everywhere
fast	LMDeploy	Maximum GPU throughput
remote	HTTP API	Lightweight client, heavy server
xpu	Intel XPU	Intel Arc GPUs

Two Components

Every mode loads two components:

1. Backbone — The transformer LLM that generates speech tokens
2. Codec — NeuCodec that decodes tokens into audio waveform

Models are auto-downloaded from HuggingFace on first use and cached at ~/.cache/huggingface/hub/.

Basic Pattern

from vieneu import Vieneu

tts = Vieneu(mode="standard")
audio = tts.infer(text="Xin chào bạn")
tts.save(audio, "output.wav")
tts.close()

Context Manager

with Vieneu() as tts:
    audio = tts.infer(text="Xin chào")
    tts.save(audio, "output.wav")
# Resources auto-released