Skip to content
Audio

fiato

a tiny speech synthesizer that can sigh

2 min readupdated

A compact English TTS (text-to-speech) model weighing in at just 4.5M parameters — 727 times smaller than its teacher model, Maya-1. It generates 24 kHz mono audio in two voices — female (F1) and male (M1) — and supports inline emotion tags right inside the text. Instead of a heavy neural vocoder, the model uses a lightweight classical synthesizer, generating audio 150 times faster than real time on a standard laptop CPU. Ideal for developers building local offline apps, games, and web services that need instant voiceovers without a GPU or the cloud.

Key features

  • Supports three expressive inline reactions right in the text: a sigh (), an exhale (), and a sharp gasp of surprise ().
  • Built on the tiny-voca architecture: combines a transformer encoder with a source-filter synthesizer (a classic pulse-and-noise generator passed through an acoustic filter) instead of a neural vocoder.
  • Ships with a ready-to-use ONNX Runtime build and runs locally right inside a browser tab without sending any text to a server.
  • Offers flexible controls for speech rate, pitch shift, and intonation pitch range.

What you can do

  • Embed it into a browser web app or extension for instant, fully client-side English text-to-speech.
  • Voice NPC dialogue in indie games or visual novels on local hardware, adding and tags for natural delivery.
  • Run it as an ultra-fast offline speech synthesizer for voice assistants on GPU-less devices.

Example

python -m fiato.infer \
  --checkpoint fiato-model/fiato.pt \
  --voice F1 \
  --text "Well,  I guess we should start over." \
  --output hello.wav
Read moreDetails
Trained via distillation on 23.7 hours of synthetic speech generated by Maya-1 (104,000 steps on a single Nvidia L40S, total budget just $18.27).
WER (word error rate measured with Whisper small.en): 0.0% on validation, 0.6% on test, and 1.0% on unseen sentences.
Speed: ~150x real-time on 4 threads of an Apple M3 Max CPU.
Limitations: US English only, two fixed voices with no voice cloning, no streaming or laughter, and foreign names may be mispronounced. Licensed under Apache-2.0.
Examples
Discuss Telegram 7 216

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 0.1 GB · ready files0.6 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.

    # cfahlgren1/fiato
    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 cfahlgren1/fiato --local-dir ~/models/fiato
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to fiato
  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 fiato (quants, finetunes, LoRAs) will show up under the bell.

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

About the model

cfahlgren1/fiato
Author
cfahlgren1
Category
Audio
Released
Posted

On NeyroLearn

Hugging Face

2167

Ask AI about this model

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

Comments

Audio

Jade 1.1

Jade 1.1