Hello, is that possible support windows or mac?

[lucasjinreal]

Many users like me wanna have a local client on windows, it would be very useful if it can support build on windows!

(I would suggest let users download onnxruntime from official themself and then link it)

I tried a little bit, the portaudio can not build on windows, and the way you writen inference onnx is not workable on windows:

model.hpp(46,78): error C2440: “<function-style-cast>”: 无法从“initializer list”转换为“Ort::Session”

Same issue:

https://github.com/microsoft/onnxruntime/issues/9001

[clockworkwhale]

I got it working in WSL (Ubuntu 22.04) under Windows 11. There seems to be little or no speed loss running it that way too.

But yeah, a native Windows version would be easier to work with. If rhasspy isn't a Windows guy then obviously it's totally fair enough for him not to bother though. Let's just hope this repo gets some more attention since that'll for sure lead to someone making a Windows fork or PR.

It's weird how little notice it's getting. I've been searching for a truly realtime TTS solution that sounds better than espeak for a long time, and this appears to finally be it. Most TTS projects that claim realtime performance are exaggerating or just lying, but inference with piper is so fast I can hardly believe it.

[lucasjinreal]

@clockworkwhale are u testing on GPU or CPU? I hope rhasspy can help support windows since there are some lib it looks hard to build on windows like portaudio, this lib even using ancient autogen and make.

[clockworkwhale]

@clockworkwhale are u testing on GPU or CPU? I hope rhasspy can help support windows since there are some lib it looks hard to build on windows like portaudio, this lib even using ancient autogen and make.

CPU (Ryzen 5950x). It's fast enough already on CPU that there's no point to using GPU, it's like 200milliseconds to generate 10 seconds of speech lol.

[Wetzel402]

I would just use WSL or Docker.

[t-mat]

Hi, Since @synesthesiam kindly merged PR #47, I think native Windows (MSVC) build becomes easier.

Here's a trivial automated build script for Win32 native version (but subset) of Piper: https://gist.github.com/t-mat/84cb1b8a828a0eee6d7db37f7ff75f70?permalink_comment_id=4545947#gistcomment-4545947

Basically, we just need to compile src/cpp/main.cpp and link it with onnruntime.lib and (patched version of) libespeak-ng.lib. Hope this helps someone who need native executable.

[lucasjinreal]

@t-mat can u help: \piper\espeak-ng-data\phontab': No such file or directory.?

[t-mat]

No, I can't. Because I do not understand how you've got that error.

[lucasjinreal]

@t-mat Please, you downloaded ng data there is no such hontab, do u get this phontab file under easpak-ng folder>?>????????

[lucasjinreal]

@t-mat Also, your espeack init didn't try to found this file?????? Are we using same espack-ng??

[lucasjinreal]

@t-mat can u tell me how to get phantab file???????

[t-mat]

As for espeak-ng, Read the README.md https://github.com/rhasspy/piper#installation

see the following lines https://github.com/rhasspy/piper/blob/283bf2851823579a67334bc61456c2c53b06bd01/Dockerfile#L16-L30

and this directory https://github.com/rhasspy/piper/tree/283bf2851823579a67334bc61456c2c53b06bd01/lib

[lucasjinreal]

@t-mat what u paste didn't make sense to me, I still didn't found phontab ....... I am using your script and I download the espack-ngpatched.zip, unziped it, didn't found this file....................

I told that am using widows, why do u still post there unrelavent content rather than directly tell me where did u downlaod phonetab file?

[t-mat]

I and anyone didn't download espeak-ng-data/phon* files. espeak-ng generates them. I'm not sure you've actually read this repository and run my script or not. But anyway, the following command generates them.

cd \path\to\espeak-ng
cd espeak-ng\src\windows
.\x64\Release\espeak-ng.exe --path=..\.. --compile-phonemes
dir ..\..\espeak-ng-data\

[notBradPitt]

@t-mat Thanks for sharing such a comprehensive build script. However, I gotta confess that I still ran into some troubles (and I reckon it's got more to do with my system than your code), so I was thinking if it'd possible for you to share your compiled build instead? That'd really save a lot of headaches here 😅

[lucasjinreal]

I tried, but the result is horrible...... is that normal? 1684592873311876800.zip

[synesthesiam]

A Windows pre-release is available here for testing: https://github.com/rhasspy/piper/releases/tag/2023.9.9-1

[clockworkwhale]

A Windows pre-release is available here for testing: https://github.com/rhasspy/piper/releases/tag/2023.9.9-1

Tested in Windows 11 on a 5950x and it's working fantastic! Many thanks for this. Speed on CPU appears to be just as insanely fast as Linux, i.e. essentially instant generation even on the larger models. ~200 ms to generate several seconds of speech.

[synesthesiam]

Thanks for testing, @clockworkwhale!

[lucasjinreal]

@clockworkwhale hello, is there any steps build on my own side?

Also, does Chinese works?

[clockworkwhale]

@clockworkwhale hello, is there any steps build on my own side?

No need to build, as synesthesiam has provided a prebuilt executable in the release he linked to.

I'm afraid I only tested English. However, Chinese voice models appear to be available on the HuggingFace page here: https://huggingface.co/rhasspy/piper-voices/tree/v1.0.0/zh

[notBradPitt]

I've tested the Windows build and it is beautiful! Piper still runs amazingly fast even when using an old CPU and the result are almost perfect. Hands down one of the best TTS I've used!

[RoxBlox3]

Hello how do i start Piper on Windows. I downloaded the Windows release and unzipped it. When i tried to open the Piper app it opens powershell and closes it immediately do i need to download something else first?

[sjc57]

What is the correct syntax for running on windows command line? It tried echo "hello" | piper --model en_US-lessac-high.onnx --output_file hello.wav but get unsupported 16-bit application popup window (I'm rinning 64 bit Windows 10) Thanks