Adds TTS to announcements made by the Captain and silicons

[Watermelon914]

Copy from #72300

About The Pull Request

TTS currently can't handle every client having their own voice, so it's been limited to announcements made by the Captain and silicon players.
In the future, if we get a GPU server, it may become a more viable option to implement every client having their own voice because GPUs are great at concurrency.

It's all handled via python http requests to a local server. Docker is required to set up, but the container can be held on a separate linux server as it's only used to process text into audio and send it back via http requests.

TTS engine used in python script:
VITS: https://arxiv.org/pdf/2106.06103
coqui: https://github.com/coqui-ai/TTS

Base image is debian.

This is all handled locally. There are no calls to a third-party service.

Why It's Good For The Game

TTS is cool.

Changelog

🆑 Watermelon914, Iamgoofball, MrStonedOne
add: Adds TTS to announcements made by the Captain and silicon players.
/:cl:

[castawaynont]

sounds awful in game

[Draggeru]

Even when the option is ticked off, I'm still hearing the TTS in game. I have no idea why.

[Mothblocks]

Can you upload an example recording of this

[TheSmallBlue]

It's supposedly testmerged on many servers but, for example, in Manuel its testmerging the old closed PR and it just. doesn't work.

[dessysalta]

When in the lobby, you can hear any TTS announcements the captain/AI makes.

[Watermelon914]

When in the lobby, you can hear any TTS announcements the captain/AI makes.

I'll look into this, this shouldn't be the case

[Watermelon914]

Can you upload an example recording of this

https://user-images.githubusercontent.com/37270891/225313437-5cdde27b-225a-40ef-a9fe-c25ab2d0e3b6.mp4
https://user-images.githubusercontent.com/37270891/225313471-de813e89-1430-44ef-9ec2-685097874c2d.mp4

I'm probably going to work on it a bit more to add more reverb since it doesn't have that intercom feel right now and it sounds like it's coming from inside your head. It needs reverb/echo and the filters make it sound robotic more than anything.
Borg voice also isn't great so I'll look into tweaking the filters for that as well. All in all, the testmerges were to just make sure it's still functional but it definitely needs polishing

[MrMelbert]

Even when the option is ticked off, I'm still hearing the TTS in game. I have no idea why.

Quite confused about this, because my TTS preferences are ticked on and I can't hear any announcements.

[MrStonedOne]

The test merge was only fixed on one of the servers it got test merged to, the other still had the version that had the inverted preferences bug.

[Mothblocks]

Can you upload an example recording of this

https://user-images.githubusercontent.com/37270891/225313437-5cdde27b-225a-40ef-a9fe-c25ab2d0e3b6.mp4 https://user-images.githubusercontent.com/37270891/225313471-de813e89-1430-44ef-9ec2-685097874c2d.mp4

I'm probably going to work on it a bit more to add more reverb since it doesn't have that intercom feel right now and it sounds like it's coming from inside your head. It needs reverb/echo and the filters make it sound robotic more than anything. Borg voice also isn't great so I'll look into tweaking the filters for that as well. All in all, the testmerges were to just make sure it's still functional but it definitely needs polishing

I know our limitations but I really wish these sounded magnitudes better, I am hesitant to call this acceptable

[Watermelon914]

I know our limitations but I really wish these sounded magnitudes better, I am hesitant to call this acceptable

These sound bad right now because there's a lack of reverb and the filters are not as good as they could be. I'm not a sound engineer so I'll need to do some more research in order to get them to a good state.
Additionally, there is the option of training my own model using the current AI VOX we have and making it silicon only, but that's going to cost money.

[Mothblocks]

They speak extremely fast and stumble over their words, we can't fix that with filters but I understand that we're extremely limited with the requirements we have for TTS

[MrStonedOne]

I'm pretty sure tempo can be messed with in the python script.

[Watermelon914]

I'm pretty sure tempo can be messed with in the python script.

You can also mess with tempo using the filters from ffmpeg. It's just that ffmpeg sucks at providing good reverb effects and you basically need another library if you want to apply reverb effects.
Though, byond already offers a way to give reverb to sound files that you send to the client, but assuming we want to play these sound files via browser to get around byond's handling of IO (to prevent clients with low latency from freezing up), then we'd need to apply the reverb to the sound file.

[Watermelon914]

Can't really find the motivation to work on this, I may end up opening it up again in the future, but I'm not really a sound engineer, I haven't got a clue on how to make it sound good and like an intercom.

[Watermelon914]

re-opening since I'm working on this again + need to respond to threads

[TheSmallBlue]

We are SO back....

[TheSmallBlue]

Damn OK, nevermind.