Suggestion: Use default windows TTS for voice output

[gyaaniguy]

First thank you for this excellent application

So windows has really awesome natural voices - available here:

https://github.com/gexgd0419/NaturalVoiceSAPIAdapter -> application to install the voice
https://github.com/gexgd0419/NaturalVoiceSAPIAdapter/wiki/Narrator-natural-voice-download-links -> links to the voices
-> follow the instructions carefully. for googlers
1. download voices -> unzip the packges to a parent folder
2. run the application -> select the parent folder with the extracted voices
3. click install 32/64
4. run C:\Windows\SysWOW64\Speech\SpeechUX\sapi.cpl as admin -> change voices in dropdown menu

There is also https://apps.microsoft.com/detail/9NLWD5DK4Q6X? but it did not really work for me. 

With this setup there is no llm or ai or online voice needed for speech to text.

Request:
Please add support to speakoflow to use default windows voice - I think its called SAPI 5 ?
Meta: Its crazy actually. All the tech is there, but disabled and deliberately being withheld

[AbhishekBarali]

Thanks for the suggestion, and thanks especially for writing out the install steps. That saved me a lot of digging, and I went and tested this properly before replying.

First thing I did was listen to the default Windows voice on my own Windows 11 machine. It was rough. Microsoft David is what most people would end up with, and on Windows 11 there is no longer any way to change the default SAPI voice, since Microsoft took it out of Control Panel and the Speech setting under Time and Language only changes the OneCore voice. So for a typical user this would mean one dated voice with no way to switch it.

The natural voices you linked are a different matter, and you are right that they sound great. What I got stuck on is how they get there. NaturalVoiceSAPIAdapter is upfront about this in its own README. It calls itself "more like a hack than a proper solution", it works by extracting encryption keys from system files, and it says it can stop working at any time after a system update. That has already happened once. The README now tells people not to install the current Narrator voice packs because they stopped working with it.

The other thing I found is that those voices are not actually running offline. The adapter converts the request to SSML and sends it to Azure AI Speech, which is why its own docs suggest making a free Azure account. So the setup is doing a cloud round trip even though it feels local.

That is the main reason I do not want to ship it as a global feature. SpeakoFlow is meant to work offline, and I am not comfortable adding an engine that quietly sends what the assistant says to a server while the setting implies it stays on your machine. On top of that, SAPI is Windows only, so it would be a feature two thirds of my users cannot use, sitting in a code path my CI does not currently build. That is a maintenance cost I would rather not take on for something I cannot promise keeps working.

None of that means the request was wrong. You pointed at a real hole. The only offline voice in SpeakoFlow today is Kokoro, and the way we ship it that is eight English voices and nothing else. If you use the app in any other language and do not want to pay for a cloud key, there is currently no voice for you at all. I want to fix that, and I would rather fix it with a proper cross platform local model than with a Windows only shim. That is what I am looking into now.

Two things that might help you today. If you want those exact Azure voices, the app already has an Azure AI Speech engine, so you can point it at your own key and get them the supported way. The free tier covers 500,000 characters a month, which is a lot of spoken replies. And since this is open source, nothing stops you keeping a local build with the SAPI patch if that is what works best on your machine.

Thanks again for taking the time to write this up properly.

[gyaaniguy]

Hey thanks for such a detailed reply !

The other thing I found is that those voices are not actually running offline. The adapter converts the request to SSML and sends it to Azure AI Speech, which is why its own docs suggest making a free Azure account. So the setup is doing a cloud round trip even though it feels local.

I decided to test it. I downloaded a random tts app > disconnected internet > tested tts with a natural voice (jenny) . It works ! So it is definitely offline
The app I tried: https://www.cross-plus-a.com/balabolka.htm

Kokoro

This didn't work for me. I got garbled sound output. As others described here: https://www.reddit.com/r/OpenWebUI/comments/1jv8szp/kokorojs_audio_issues_in_chrome/
Perhaps I can help you diagnose this ? Please let me know what i can do to help..

That is a maintenance cost I would rather not take on for something I cannot promise keeps working.
That is most understandable.

Thanks again