Read out loud --- chatgpt?

[MedBooster]

In the past we've discussed dictation (speech to text)

What about the Read-out-loud-ttool (text to speech)?

Any good options or way to use chatgpt for this? For a more natural sounding voice.

Maybe without using a third-party program, from within PDF-XCE itself?


Currently my favourite is in Loki@99's threads
viewtopic.php?t=42657&hilit=read+out+loud
However, this is not AI-powered but just a regular text-to-speech engine by Microsoft. (free)

[Daniel - PDF-XChange]

Hello, MedBooster

To my knowledge, ChatGPT specifically is the language model, not a Voice model. Though many services that use ChatGPT may have a vocal component, my understanding is GPT uses a highly specialized Voice synthesis model, which is extremely expensive to run, and that simply parses the GPT's (text/conversation structuring) "Language logic".

In any case, such functionality would most likely require one of two things. Either;

1. We directly integrate an AI assistant to the software, which is able to read and parse your document content in full, so it can determine phrasing cadence to allow for more natural speaking (This is a very complex process, and to my knowledge, most locally run services do not offer that kind of synthesis, because the average users PC is not powerful enough to handle it).
2. We offload the text that is to be read, to something external on the system, allowing it to do the same, without us needing to directly offer AI integration.

(There are of course other options, but the above are my personal predictions based on what I know of current AI technology/use)

As you can probably imagine, such functionality is complicated if it cannot remain entirely local because of the potential security risks. arbitrarily sending your document data to an unrelated 3rd party could be a major issue, and even if we go through the complex process of offering something of the like in the future, it would likely require an external user agreement, and possibly your own licensing/login through the third party service used.

In short, I would not get my hopes up for something from us on this anytime soon. I might suggest that you look for a voice which works with the inbuilt Windows Voice API (which we currently use and works locally), that is able to offer more natural sounding speech, without the "AI voice synthesis" functions that allow for those engines to offer notably more "human" sounding speech.

Alternatively, if you want GPT specifically to read a PDF out to you right now, you may be able to upload the file to ChatGPT directly, or export it as text, and do the same, asking it to read it out to you that way.

Kind regards,

[MedBooster]

Like you said, it is expensive to run, and even the plus (paid) version of ChatGPT has a daily limit, which I seem to be reaching quite frequently...

[Daniel - PDF-XChange]

Hello, MedBooster

When I said expensive, I meant more in terms of hardware resources. From what research I did, it looks like the server hardware required to run these voice synthesis systems are... well above what the the most invested Gamers/Graphic Designers would have on hand, let along the average everyday PDF software user. So running it "locally" is not likely an option to begin with, at least with modern hardware. Maybe in another decade...

That only leave the online "use something that exists" option, which would effectively mean you would have the same costs you do now, if for example, we chose to connect with GPT specifically. You would need to sign into their system, and use their credits in conjunction. (Which would be expensive in a more literal sense)

Kind regards,