Modelpublic

tmdarkbr/echo-tts-gguf

sourceHugging Facecc-by-nc-sa-4.0updated 5mo agoView on Hugging Face

3likes70downloads

Model Card

Echo-TTS GGUF + ONNX

Converted model files for the Echo-TTS C++ inference engine.

This repo contains:

—`echo-dit.gguf` - EchoDiT diffusion transformer weights (F16 GGUF, best quality)
—`echo-dit-q8.gguf` - EchoDiT diffusion transformer weights (Q8_0 GGUF, near-lossless, lower VRAM)
—`onnx/` - S1-DAC autoencoder (ONNX format) for audio encoding/decoding

Converted from Echo-TTS by Jordan Darefsky.

Download

bash

# F16 GGUF model (5.6 GB) — best quality
huggingface-cli download tmdarkbr/echo-tts-gguf echo-dit.gguf --local-dir .

# Q8_0 GGUF model (3.3 GB) — near-lossless, lower VRAM
huggingface-cli download tmdarkbr/echo-tts-gguf echo-dit-q8.gguf --local-dir .

# ONNX DAC models
huggingface-cli download tmdarkbr/echo-tts-gguf onnx/ --local-dir onnx_models

Usage

bash

echo-tts \
  --model echo-dit.gguf \
  --speaker speaker.wav \
  --dac-encoder onnx_models/dac_encoder.onnx \
  --dac-decoder onnx_models/dac_decoder.onnx \
  --text "[S1] Hello world." \
  --output output.wav

Use echo-dit-q8.gguf instead of echo-dit.gguf for lower VRAM usage with near-lossless quality.

Files

File	Size	Description
`echo-dit.gguf`	5.6 GB	EchoDiT transformer (2048-dim, 24 layers, F16)
`echo-dit-q8.gguf`	3.3 GB	EchoDiT transformer (2048-dim, 24 layers, Q8_0)
`onnx/dac_encoder.onnx`	1.5 MB	S1-DAC encoder graph
`onnx/dac_encoder.onnx.data`	796 MB	S1-DAC encoder weights
`onnx/dac_decoder.onnx`	0.9 MB	S1-DAC decoder graph
`onnx/dac_decoder.onnx.data`	704 MB	S1-DAC decoder weights

License

—Model weights: CC-BY-NC-SA-4.0 (same as original Echo-TTS)
—Audio outputs: CC-BY-NC-SA-4.0 (due to Fish Speech S1-DAC dependency)

Citation

bibtex

@misc{darefsky2025echo,
    author = {Darefsky, Jordan},
    title = {Echo-TTS},
    year = {2025},
    url = {https://jordandarefsky.com/blog/2025/echo/}
}