Marcel Timm, RhinoDevel, 2025

mt_tts is a C++ library for Linux and Windows that offers a pure C interface to the awesome text-to-speech system called Piper by Michael Hansen.

mt_tts supports:

• Text to wave file.
• Text to raw audio samples.
• Text to raw audio stream-to-stream conversion.

Take a look at the example showing a simple Speech-To-Text, Large-Language-Model, Text-To-Speech pipeline via mt_stt, mt_llm and mt_tts!

Clone the mt_tts repository:

git clone https://github.com/RhinoDevel/mt_tts.git

Enter the created folder:

cd mt_tts

Get the Piper submodule content:

git submodule update --init --recursive

If you want to build Piper in debug instead of release mode, you need to modify the following in the file CMakeLists.txt:

fmt and spdlog under target_link_libraries( (two times) into fmtd and spdlogd.

No details for Linux here, yet, but you can take a look at the Windows instructions below and at the Makefile.

• Open Visual Studio developer commandline.
• Go to the Piper submodule folder: cd mt_tts\piper
• Create build folder: mkdir build
• Enter build folder: cd build
• Prepare the build: cmake ..
• Start the build (this will also download stuff Piper needs from the internet): cmake --build . --config Release (or cmake --build . for debug mode)

Copy the following (from different output directories in the build folder) into a new folder:

• build\pi\share\espeak-ng-data (the whole folder)
• build\pi\bin\espeak-ng.dll
• build\pi\bin\piper_phonemize.dll
• build\pi\lib\onnxruntime.dll
• build\Release\piper.exe (or build\Debug\piper.exe)

Download a voice and its configuration, e.g. one for speech output in German language by Thorsten Müller:

• Model file (ONNX)
• Model configuration file (JSON)

Store these files in the same, new folder.

echo "Ich bin ein Mensch, Du auch?" | piper --model de_DE-thorsten-high.onnx --config de_DE-thorsten-high.onnx.json --debug --output_file test.wav

Play back the WAV file:

ffplay.exe test.wav

(ffmpeg parameters may not be optimal, in this example):

echo "Hallo, ich bin kein Mensch, was man auch einigermaßen leicht heraushören kann, meinst Du nicht auch? Trotzdem ein tolles TTS-System!" | piper --model de_DE-thorsten-high.onnx --config de_DE-thorsten-high.onnx.json --output_raw | ffplay.exe -f s16le -ar 22050 -

• Open solution mt_tts.sln with Visual Studio (tested with 2022).
• Compile in release or debug mode.

• Get the DLL and LIB files resulting from the build, e.g. for release mode x64\Release\mt_tts.dll and x64\Release\mt_tts.lib, copy them to a new folder.
• Also copy the file mt_tts\mt_tts.h to that new folder.
• Copy the following stuff from Piper to the new folder, too:
  • build\pi\share\espeak-ng-data (the whole folder)
  • build\pi\bin\espeak-ng.dll
  • build\pi\bin\piper_phonemize.dll
  • build\pi\lib\onnxruntime.dll
• Also copy a voice model file and its configuration file to the same new folder.
• Open x64 Native Tools Command Prompt for VS 2022 commandline.
• Go to the new folder and create a file main.c with the following code:

#include "mt_tts.h"

int main()
{
    mt_tts_reinit("de_DE-thorsten-high.onnx", "de_DE-thorsten-high.onnx.json");

    mt_tts_to_wav_file(
        "Hallo, nun testen wir dieses kleine Hilfsmodul.", "output.wav");

    mt_tts_deinit();

    return 0;
}

• Compile via cl main.c mt_tts.lib.
• Run main.exe, which will create the WAV file output.wav.