Building a single .exe with this fork — is it possible?

[HardCodeDev777]

Hi! I used the original Coqui TTS for some time, and surprisingly, it worked perfectly.
Now I need to make it a single .exe. I tried using PyInstaller and wasted a lot of time with countless errors like:

FileNotFoundError: [Errno 2] No such file or directory: '...\\AppData\\Local\\Temp\\_MEI296122\\TTS\\VERSION'

FileNotFoundError: [Errno 2] No such file or directory: '...\\AppData\\Local\\Temp\\_MEI39722\\trainer\\VERSION'

ModuleNotFoundError: No module named 'scipy._cyutility'

Of course, I can try --add-data "...\Project\.venv\Lib\site-packages\TTS;TTS" and fix other missing modules manually(and I tried a lot), but that would take forever to fix all errors. Another problem is that I’m not very familiar with Python—I mostly rely on ChatGPT.

So, I want to ask: is it possible to make a .exe with this fork? If it is, I will try to do it.

[eginhard]

I'm not familiar with PyInstaller, but perhaps you could share the full command you're running so that somebody else would be able to help?

Otherwise the simplest way to install the Coqui command-line interface (CLI) as a stand-alone application would be to install uv following the instructions at https://docs.astral.sh/uv/getting-started/installation. Then run:

uv tool install coqui-tts

Then run the CLI as usual: https://coqui-tts.readthedocs.io/en/latest/inference.html#command-line-interface-tts

[HardCodeDev777]

That sounds really cool, thank you so much for the quick reply! I want to make it without Python - no matter it's .exe or CLI , so I’ll definitely give it a try with uv as soon as I can.

[eginhard]

Yes, uv will automatically install Python behind the scenes if it's not available yet.

[HardCodeDev777]

If it works for me, I’ll close this discussion.

[ROBERT-MCDOWELL]

FYI create a .exe without microsoft certification windows will consider it as a potential threat and will be put directly in quarantine.....

[HardCodeDev777]

I don’t think that’s the issue here – Microsoft Defender hasn’t flagged or quarantined anything. The .exe runs, but the problem seems to be specifically with missing modules. I also don’t think Defender would remove PyInstaller’s temporary files.

[ROBERT-MCDOWELL]

I knw it's not the problem here. I just advised you if you planned to share your .exe with other users.

[HardCodeDev777]

Thanks! Didn't really know about it.

[HardCodeDev777]

I tried using uv as suggested, but I’m running into the following issue:

ValueError: You passed a file to checkpoint_dir=. Use checkpoint_path=...\TTSModel\model.pth instead.

I also tried using --checkpoint_path, but the CLI reports that such an argument doesn’t exist.

I want to clarify that I’m using an XTTS multi-speaker model, and I need it to take the voice from my own .wav file. I couldn’t find this option under the Multi-speaker models section in the CLI documentation.

When I was using a Python script, I could simply pass the .wav file like this:

tts.tts_to_file(
    text=text,
    speaker_wav=speaker_file,
    language=lang,
    file_path=buf
)

Is there a way to achieve the same functionality using the CLI?

I run the following command:

tts --text "Test audio" --model_path "...\TTSModel\model.pth" --config_path "...\TTSModel\config.json" --source_wav "...\Voices\voice.wav" --out_path "...\result.wav"

I guess there's no --source_wav argument for XTTS multi-speaker model.

[eginhard]

Yes, the error message is not very good when using the CLI there. Pass only the path ...\TTSModel\ instead and leave out the model file name. But if you're just using the original Coqui model, not a fine-tuned version I'd recommend to instead run it with the model name as described here: https://coqui-tts.readthedocs.io/en/latest/models/xtts.html#single-reference

[eginhard]

No, you need to use --speaker_wav just like in Python, see: https://coqui-tts.readthedocs.io/en/latest/models/xtts.html#single-reference

[HardCodeDev777]

YES! It finally works!
The only "problem" I have - when --use_cuda I get this error:

raise AssertionError("Torch not compiled with CUDA enabled") AssertionError: Torch not compiled with CUDA enabled

Without --use_cuda everything works perfectly, but I'd like to use cuda

[eginhard]

For Windows, PyPI provides the CPU version of Pytorch. You could try the following to install the GPU version from the Pytorch index instead (not sure if this will work):

uv tool uninstall coqui-tts
uv tool install --index https://download.pytorch.org/whl/cu128 coqui-tts

You might have to adjust that URL depending on your Cuda version, for the available options see https://pytorch.org/get-started/locally/

[HardCodeDev777]

Ok, I'll try it. Anyway, it works!
Thanks a lot for maintaining this fork and for helping me figure things out! It really helped me, and I truly appreciate the effort you put into keeping this project alive.

[HardCodeDev777]

Also noticed another problem - it works with --model_name tts_models/multilingual/multi-dataset/xtts_v2, but not with --model_path "...\TTSModel\" (I get tts: error: unrecognized arguments error). In my case, it's important to keep model in specific folder.

[HardCodeDev777]

Fixed problem with --model_path - "...\TTSModel", not "...\TTSModel\"😅
But still don't know why --use_cuda doesn't work.