Skip to content
Audio

Jade 1.1

a compact TTS model built to run right on your phone

2 min readupdated

Jade 1.1 is a compact multilingual TTS (text-to-speech) model from ETM-Network designed to run locally on low-power hardware. It converts raw text directly into 44.1 kHz audio without a separate phonemizer, supporting 23 languages—including Russian—and 9 selectable voices. Under the hood, a 12-block diffusion transformer (84M parameters) generates latent representations for a Supertonic 3 decoder (25M parameters), bringing the total to 109M parameters. Unlike heavy cloud solutions, the base quantized build weighs in at around 150 MB and runs 4× faster than real time even on a mid-range smartphone CPU. It is a solid fit for mobile app developers, offline bots, and voice interfaces.

Key features

  • Supports 23 languages (including Russian) and 9 voices (5 female, 4 male), each able to speak any of the supported languages.
  • Available in ONNX and MLX formats with q8 (150 MB), q4 (81 MB), and q3 (65 MB) quantization, alongside the original float32 weights (437 MB).
  • Runs ~4× faster than real time on a Snapdragon 778G+ mobile CPU, ~7× on a desktop Intel i9-13900KF, and up to ~250× on an RTX 4090.
  • Distilled for low-step generation (2 to 32 steps, 12 by default), letting you fine-tune the trade-off between inference speed and prosody stability.

What you can do

  • Build offline text-to-speech and accessibility features into Android apps or local services without sending data to the cloud.
  • Narrate long articles and audiobooks in a neutral announcer tone using the q8 build at 12–16 steps.
  • Spin up a fast voice responder or assistant prototype on an Apple Silicon Mac using MLX.

Example

from etm_tts import TTS

tts = TTS()  # скачивает onnx/quants/q8 (~150 МБ)
tts.speak("Hello! It's nice to finally meet you.").save("hello.wav")
tts.speak("Guten Morgen, wie geht es dir?", voice="Henry", lang="de").save("morgen.wav")
Read moreDetails
On a 181-phrase benchmark (8 per language), the model scored a 1.1% Character Error Rate via Whisper transcription and 0.818 speaker similarity (WavLM) in its default preset (q8, 12 steps).
Requires ONNX Runtime 1.28+ (use onnxruntime-gpu for CUDA) or etm-tts[mlx] on Apple Silicon; a dedicated Android library is also available.
Limitations: language must be specified manually; numbers, dates, and abbreviations are normalized only in English and German (for Russian and other languages, numbers must be written out in words); no emotion control; occasional pitch jumps may occur on long texts at low step counts (q4/q3).
License: BigScience OpenRAIL-M (the etm-tts runtime code is Apache 2.0).
Examples
Discuss Telegram 0

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 0.3 GB · ready files0.8 GB
  • FP8 / INT8recommendedweights 0.2 GB0.7 GB
  • 4-bitweights 0.1 GB0.6 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.

    # ETM-Network/jade-1.1
    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 ETM-Network/jade-1.1 --local-dir ~/models/jade-1.1
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to jade-1.1
  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 jade-1.1 (quants, finetunes, LoRAs) will show up under the bell.

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

About the model

ETM-Network/jade-1.1
Author
ETM-Network
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

fiato

fiato