[Code Bounty] Spaceman TTS: Definitive Edition
Posted: Wed Jul 06, 2022 9:47 pm
Right, so, remember this PR? I think it was sick, other people do too, so I'm willing to pay 200 dollars for it to done in a more merge-able matter.
This is quite the request, and by that i mean that its an utter pain. I don't expect this code bounty to be claimed any time soon, maybe in the near future when a new expressive open source tts solution arrives.
If you for some reason are interested, here are some guidelines:
[*] You cannot use another service (like what the original PR did with NovelAI) (The delay and cost wouldn't be worth it, besides the fact that maintainers will simply not merge it)
[*] The synthesized speech should be as expressive as the one provoked by the above PR. Here are some examples.
[*] Each player's speech should be customizable to a degree in which no two players should sound alike. This can be done via simple pitch or tremble modifications or using a seed system like the one used in the NovelAI PR.
[*] It should be turned off by default and enabled in the options menu. In the same options menu should be the speech customization options I mentioned above.
[*] Delay between sending a message and it being synthesized can not be null, but it should be as minimal as possible.
If all the guidelines are met, you theoretically shouldn't have any problem with merging. Mothblocks has said that they want TTS to be a thing, so I doubt there'll be much opposition.
The problem is, yknow, actually meeting the guidelines. As of writing this the only possible way this could be done is via machine learning. There are some github repositories with machine learning models that claim to be expressive, however they have to be trained, so I can't really test them.
Any questions you may leave them down here and I'll edit the post as I see fit. If you want to message me, my discord is SmallBlue#9038. To be able to message me you should be on the tgstation server or coderbus.
Payment (as of right now) is 200$, paid via paypal, on merge.
This is quite the request, and by that i mean that its an utter pain. I don't expect this code bounty to be claimed any time soon, maybe in the near future when a new expressive open source tts solution arrives.
If you for some reason are interested, here are some guidelines:
[*] You cannot use another service (like what the original PR did with NovelAI) (The delay and cost wouldn't be worth it, besides the fact that maintainers will simply not merge it)
[*] The synthesized speech should be as expressive as the one provoked by the above PR. Here are some examples.
[*] Each player's speech should be customizable to a degree in which no two players should sound alike. This can be done via simple pitch or tremble modifications or using a seed system like the one used in the NovelAI PR.
[*] It should be turned off by default and enabled in the options menu. In the same options menu should be the speech customization options I mentioned above.
[*] Delay between sending a message and it being synthesized can not be null, but it should be as minimal as possible.
If all the guidelines are met, you theoretically shouldn't have any problem with merging. Mothblocks has said that they want TTS to be a thing, so I doubt there'll be much opposition.
The problem is, yknow, actually meeting the guidelines. As of writing this the only possible way this could be done is via machine learning. There are some github repositories with machine learning models that claim to be expressive, however they have to be trained, so I can't really test them.
Any questions you may leave them down here and I'll edit the post as I see fit. If you want to message me, my discord is SmallBlue#9038. To be able to message me you should be on the tgstation server or coderbus.
Payment (as of right now) is 200$, paid via paypal, on merge.