Any solutions for C++?

[0_x_ol_lo_x_0 (firefox_email@fastmail.com)]

I have attempted and failed to find a solution for TTS in C++ that sounds as good as the demos I have heard from Mozilla TTS. Currently my only choices are mimic1 or Windows SAPI, neither of which sound good enough for what I want to accomplish.

Cloud solutions are not an option, and I cannot use python to interface with C++ or vice versa (this is for an Unreal Engine 4 project – and I need to run TTS, not train models with C++).

I would love to know if anyone has suggestions for the above use case (I’m starting to think I’m out of luck at the moment).

[dkreutz (Dominik)]

Don‘t know if this actually works in your case, but maybe worth a look:

pradyun/Py2C

Or this:

Nuitka/Nuitka

Major problem will be how libraries like Torch, Numpy, etc. will be handled (some of them are written in C/C++ though).

[nmstoker (Neil Stoker)]

Not sure if you’re considering this as ruled out by “cloud” but what about running the TTS server locally and calling it from C++?

I do recall quite some time back that there was someone in the forum who mentioned running the models on C (IIRC) as they’d adapted it to work themselves but I don’t think they released their code so probably isn’t of use to you.

[nmstoker (Neil Stoker)]

Here’s the post I was thinking of (was C++ actually)

C++ Implementation of synthesizer for the Tacotron model based on OpenCV capable of running on mobile devices

We’ve managed to create a C++ implementation of the Tacotron multi-speaker embedding model based on OpenCV which runs near real-time or faster on contemporary mobile devices. The training is done in Python using the original MozilaTTS implementation and then the data is converted with python script into easier-to-read-from-CPP data format. The data size of the converted Tacotron model is approx. 28MB of binary data; in our latest pre-trained model we have two voices simultaneously (male and fe…

[ZDisket]

I was wondering this myself while looking for a way to integrate Mozilla (now Coqui) TTS models into my Qt program. TorchScript is broken, and so is exporting to Tensorflow SavedModel (how I currently use another repo’s models) after running the Tensorflow conversion code: it exports without errors, but the resultant model is incapable of doing anything but outputting noise even when fed the very same input that generated perfectly fine speech from Python. ONNX? Also impossible.

There seems to be no hope for C++ users.