search

joshjaysalazar / IntelliCaster

An AI race commentator for iRacing
GNU General Public License v3.0
3 stars 1 forks source link

Other TTS And Text Generation Languages #92

Open Skutela32 opened 3 months ago

Skutela32 commented 3 months ago

Is your feature request related to a problem? Please describe. Using for longer sessions will cost a significant amount towards the TTS software

Describe the solution you'd like An alternative option to use services like AWS or Google Cloud

Describe alternatives you've considered Google Cloud and AWS

Skutela32 commented 3 months ago

I will help where i can within this

joshjaysalazar commented 3 months ago

I don't disagree with this, but it should also have a warning about quality. My goal with this software was for it to sound as human as possible. The cheaper you go with TTS, the more obvious it is that it's a computer talking, and I don't want people to judge the software on cheap sounding voices.

joshjaysalazar commented 3 months ago

But, to be clear, I do think this is a feature worth adding. I was hoping OpenAI's newest TTS features would be available by API by now, because they sound pretty good and are far cheaper to run, but we're still waiting on the newest 4o version (the old ones are available, but sound lifeless). As soon as the 4o TTS is available in the API, it will definitely be an option.

So perhaps a new drop-down is in order to choose your TTS preference, along with maybe tooltips or something to explain the difference in both quality and cost.

Skutela32 commented 3 months ago

Yea drop downs for the gpt version along with the tts version. I only suggested it as i was planning to use this for some commentary on youtube videos but doing multiple 30 min videos with commentary a week is gonna cost more then its worth.

joshjaysalazar commented 3 months ago

Fair, for sure. Definitely a good idea to have cheaper options in the settings. Maybe even going so far as to implement local options if the user's GPU is up to the task, I dunno.

You'll want to wait until a few more commentary features are implemented as well, I think. There's still a big awkward silence at the beginning, no indication of when the race ends, etc. 😂

Skutela32 commented 3 months ago

You never know tho, if its gets popular then could possably get some additional help from people who know a bit more. Or even you know turn it into a little program you could offer subscriptions for (to cover api costs)

joshjaysalazar commented 3 months ago

You never know tho, if its gets popular then could possably get some additional help from people who know a bit more.

Hopefully! I still consider this a very early prototype, but hopefully people, especially league admins, will find it helpful when it's more complete.

Or even you know turn it into a little program you could offer subscriptions for (to cover api costs)

That's definitely an interesting idea, and also something to be considered way down the road.

Skutela32 commented 3 months ago

That's definitely an interesting idea, and also something to be considered way down the road.

Demonstrate an early (but functional) product can get people interested to see how it goes along the way

joshjaysalazar commented 3 months ago

Just wanted to note, it's worth keeping an eye on OpenAI's most recent updates. In particular, they released GPT-4o mini, obviously just text generation, not specifically TTS, but still worth noting. It's cheaper than 3.5-turbo, but still has the significantly bigger context window.

Still crossing my fingers for API access to the updated OpenAI TTS as well. If it stays as cheap as it has been but adds the ability to control emotions better, that will definitely need to be added as an option. ElevenLabs is still the top service for believable AI TTS, but OpenAI will be a close second I think once that's available.