mmwillet2/Dia_GGUF

Purpose

The purpose of this repository is to store various TTS.cpp compatible GGUF encoded model files for the Dia model.

Model Types

Currently the model is supported with 4-bit, 5-bit, 8-bit, F16bit and F32bit quantization / precision and all modes are supported with F16 and F32 bit precision DAC. Dia.gguf is the non-quantized 32 bit floating point version, Dia_Q4.gguf, Dia_Q5.gguf, Dia_Q8.gguf and Dia_F16.gguf are the 4bit, 5bit, 8bit and 16bit quantized versions respectively, and all versions with the prefix _DAC_F16.gguf are encoded with a 16bit version of the DAC audio encoder.

Dia

This page only contains the GGUF encoded model files of the original Dia model. For the original model please see the repository here.

How to use

See the github repo here for more information general usage.

To compile TTS.cpp simple git clone and then run the the following in the repository's directory to compile (cmake is required):

cmake -B build                                           
cmake --build build --config Release

After compilation is complete you can download a model file generate speech to a file from the same directory like so:

build/bin/cli --model-path /model/path/to/downloaded_gguf_file.gguf --prompt "I am saying some words" --save-path /tmp/test.wav