Skip to content
Audio

tiny-voca Kiswahili

a lightweight offline Swahili text-to-speech model

2 min readupdated

This compact text-to-speech (TTS) model for Swahili has just 4.54 million parameters, built to run directly on low-end mobile devices like Android smartphones without cloud connectivity. Based on the tiny-voca architecture (a source-filter vocoder with a duration block), the model is exported to the ONNX format and takes up about 21.4 MB. It generates high-quality Swahili speech supporting seven fixed voices. The solution is ideal for mobile app and screen reader developers who need a fully autonomous voice engine with minimal RAM usage.

Key features

  • Mobile optimization: runs stably on budget Android smartphones with minimal RAM consumption (71–75 MB).
  • High speed: CPU speech synthesis with a real-time factor (RTF) between 0.16 and 0.21.
  • Precise duration alignment: outputs phoneme durations directly during generation, making it easy to implement synchronized text highlighting.
  • Compact size: weighs only 21.4 MB in ONNX format and operates without heavy cloud servers.

What you can do

  • Mobile app integration: embed offline text voiceovers into local apps and readers for Swahili-speaking users.
  • Text-highlighting systems: build interactive reading tools for constitutions or educational materials with real-time visual highlighting of spoken words.
  • Autonomous voice bots: deploy a lightweight audio generation system on inexpensive server CPUs or local devices without a GPU.

Example

An example of running model inference using the ONNX Runtime library in Python:

import numpy as np, onnxruntime as ort
sess = ort.InferenceSession("swahili.onnx", providers=["CPUExecutionProvider"])
audio, durations = sess.run(None, {
    "tokens": np.array([tokens], dtype=np.int64),
    "stress": np.array([stress], dtype=np.int64),
    "voice": np.array([0], dtype=np.int64),
    "speed": np.array([1.0], dtype=np.float32),
    "pitch_shift": np.array([0.0], dtype=np.float32),
    "pitch_range": np.array([1.0], dtype=np.float32),
})
Read moreDetails
Architecture: source-filter vocoder with duration block, 4 encoder slots
Parameters and size: 4.54M parameters, 21.4 MB (ONNX, fp32)
Sample rate: 24 kHz
Limitations: up to 512 tokens and 32 seconds of audio per call; numbers and abbreviations require preliminary text expansion
License: CC-BY-SA-4.0 (weights and WAXAL dataset)
Discuss Telegram 0 20

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.

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

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

About the model

adoamesh/tiny-voca-swahili
Author
adoamesh
Category
Audio
Released
Posted

On NeyroLearn

Hugging Face

200

Ask AI about this model

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

Comments