Skip to content
Audio

tiny-voca-zh

an ultra-compact Chinese text-to-speech model

2 min read

Based on the tiny-voca architecture, this model features an acoustic system with a built-in waveform synthesizer. Weighing just 4.5 million parameters, it takes Chinese text as input and outputs ready-to-use 24 kHz audio without requiring a separate neural vocoder. The repository provides two fixed voices (female F and male M), while text preprocessing is handled by a built-in frontend based on the cn2an, jieba, and pypinyin libraries. This project is ideal for developers who need fast, local Chinese text-to-speech generation on low-end hardware.

Key features

  • Minimal size: only 4.5M parameters and a weights file size of 18.2 MB.
  • Two voices: built-in presets for female (F) and male (M) voices.
  • Direct output: generation of 24 kHz mono audio without an intermediate vocoder.
  • Ready-to-use frontend: built-in inventory of 115 characters for processing Chinese phonetics and homophones.

What you can do

  • Script integration: run speech generation for Chinese texts via Python code or the command line.
  • Fine-tuning: adjust speech speed, shift, and pitch range using command line parameters.
  • Local deployment: deploy a lightweight text-to-speech service on local hardware with minimal VRAM requirements.

Example

import os
import sys
import soundfile as sf

sys.path.insert(0, "src")
os.environ["TINY_VOCA_FRONTEND"] = "zh"

from tiny_voca.infer import load_model, speak_text, limit_peak

model = load_model("tiny-voca-zh.pt", "cuda")
wave, _ = speak_text(model, "她在笔记本上写下一行字:明天会更好。", "F")
sf.write("out.wav", limit_peak(wave).numpy(), model.config.sample_rate, subtype="FLOAT")
Read moreDetails
CER (character error rate according to Paraformer ASR) benchmarks: general student — 2.16%, voice F — 2.52%, voice M — 1.80%, teacher (Qwen3-TTS-12Hz-1.7B) — 1.40%.
Training: trained from scratch on 22,298 synthesized sentences (~23.6 hours).
Limitations: supports reading text only with fixed voices, does not control emotions and speech style.
License: Apache-2.0.
Discuss Telegram 0

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 0.0 GB · ready files0.5 GB
  • FP8 / INT8recommendedweights 0.0 GB0.5 GB
  • 4-bitweights 0.0 GB0.5 GB

Tip: go 8-bit.FP8 / INT8 is within about a percent of full precision in quality, takes half the memory and runs faster on RTX 40/50 and H100. Full precision only pays off for fine-tuning.

What to get for this model

Massed ComputeOpen Massed Compute

Your own GPU: FP8 / INT8 — from RTX 2060 · BF16 — from RTX 2060

Massed Compute hourly prices, November 2026. The discounted price applies with the coupon at checkout.

Estimate: weights plus working memory. Long videos, high resolutions and long contexts need more.

Run it on Massed Compute

The script downloads the model to the rented machine. How to run it is in the model card.

  1. 1. Rent a machine with a fitting GPU

    Open Massed Compute
  2. 2. Paste the script into the machine's terminal

    Connect over SSH (the Massed Compute panel shows the command) or open a terminal on the machine's desktop.

    HF_TOKEN / CIVITAI_TOKENThe script starts with token lines: paste yours between the quotes. The Hugging Face token is for gated models, the Civitai one for Civitai models.

    # songLucasXXX/tiny-voca-zh
    mkdir -p ~/models
    curl -LsSf https://astral.sh/uv/install.sh | sh
    source "$HOME/.local/bin/env"
    uv tool install "huggingface_hub[cli]"
    hf download songLucasXXX/tiny-voca-zh --local-dir ~/models/tiny-voca-zh
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to tiny-voca-zh
  3. 3. Use it

    The model is in ~/models. How to run it is in the model card (link above).

Stop the machine in the panel when you are done: it bills by the minute while it runs.

How to speed it up

Picked for this model type. Choose your hardware above to see what fits you exactly.

  • FlashAttention 44.0.0b40from November 18, 2026new release

    fastest exact attention on B200

    Kernels for Blackwell (beta). Exact attention without quantisation — fine for everything, LLMs included.

    How to enable: pip install flash-attn-4; vLLM and SGLang pick it up themselves.

    Earlier generations (2)
    • FlashAttention 3 — The Hopper (H100/H200) version with FP8.
    • FlashAttention 2 — The standard for RTX 30/40/50 and A100: exact attention without the memory cost of long contexts.
  • torch.compile2.15.0from October 28, 2026new release

    10–30% for free

    Compiles the model for your GPU. Slower first run, faster afterwards.

    How to enable: model = torch.compile(model); in ComfyUI the TorchCompileModel node.

Model family

New posts about versions of tiny-voca-zh (quants, finetunes, LoRAs) will show up under the bell.

No known relatives of this model yet. All derivatives on Hugging Face

About the model

songLucasXXX/tiny-voca-zh
Author
songLucasXXX
Category
Audio
Released
Posted

On NeyroLearn

Hugging Face

00

Ask AI about this model

Answered by Gemini 4 · the chat is kept only in this tab

Comments