Skip to content
Audio

mg-voca: Malagasy mobile TTS

2 min read

The mg-voca model handles text-to-speech tasks for the Malagasy language, producing 24 kHz audio. It supports two voices (female and male) and weighs only 21 MB in ONNX format, enabling local operation on a CPU, in a browser, or on a smartphone without an internet connection. The model architecture is based on the miniature tiny-voca scheme and includes a transformer-based encoder, a duration predictor, and a waveform generator without a separate vocoder. It is well-suited for developers of offline applications, voice assistants, or pipelines paired with speech recognition (STT) systems for the Malagasy language.

Key features

  • Compact size: the model weighs only 21 MB, allowing it to run on mobile devices.
  • High execution speed: the median sentence generation time is about 146 ms on two CPU threads.
  • Two voices: built-in female and male voices trained on data from the Afrivoice Malagasy project.
  • Easy setup: operation via ONNX Runtime without heavy dependencies like PyTorch.

What you can do

  • Build an offline assistant: integrate the model into a mobile or web application to read out responses without internet access.
  • Pair with STT: combine it with a speech recognition model for two-way voice input and output.
  • Synthesize speech lines: voice short commands, notifications, and status messages in a Malagasy language interface.

Example

pip install onnxruntime numpy soundfile huggingface_hub
hf download darioooooo0o/malagasy-tts-mgvoca --local-dir mg-tts
python mg-tts/examples/quickstart_tts.py mg-tts "Manao ahoana, inona no vaovao?" hello.wav F
Read moreDetails
• Parameters: about 4.5 million parameters.
• Performance: median of 146 ms per sentence (105x faster than real-time) on 2 CPU threads.
• Quality: naturalness rating of 3.5 out of 5 (according to a single tester).
• License: Apache-2.0 for code and model weights; training audio data is distributed under the CC-BY-4.0 license.
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.

    # darioooooo0o/malagasy-tts-mgvoca
    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 darioooooo0o/malagasy-tts-mgvoca --local-dir ~/models/malagasy-tts-mgvoca
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to malagasy-tts-mgvoca
  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 malagasy-tts-mgvoca (quants, finetunes, LoRAs) will show up under the bell.

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

About the model

darioooooo0o/malagasy-tts-mgvoca
Author
darioooooo0o
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

Audio

pingo-ptbr

pingo-ptbr