A New State-of-the-Art Text-to-Speech Program - Install & Testing MaskGCT

Links referenced in the video:
MaskGCT Demo Page - maskgct.github...
Phonemizer fix - github.com/ope...
Laptop that I use:
Lenovo Thinkpad P1 Gen 7 - bit.ly/4haIt46
Hardware for my PC:
Graphics Card - amzn.to/3pcREux
CPU - amzn.to/43O66Ir
Cooler - amzn.to/3p98TwX
RAM - amzn.to/3NBAsIq
SSD Storage - amzn.to/42NgMFR
Power Supply (PSU) - amzn.to/430bIhy
PC Case - amzn.to/447499T
Mother Board - amzn.to/3CziMXI
Alternative prebuilds to my PC:
Corsair Vengeance i7400 - amzn.to/3p64r22
MSI MPG Velox - amzn.to/42MnJHl
Cheapest and PC recommended:
Cyberpower 3060 - amzn.to/3XjtZoP
Come join The Learning Journey!
Discord - / discord
Github - github.com/Jar...
TikTok - / jarodsjourney
If you found anything helpful, please consider supporting me and the content I am trying to produce!
www.buymeacoff...

[@Jarods_Journey]

Timestamps: 2:00 - Running Through Install 11:05 - Start of Testing the TTS

[@dathuynh-l4k]

Can gtx 1660 run this program?

[@SuperSigma69]

Brokie get to work n get a 4090

[@rons96]

for maskGCT i found that best results come with short reference audios. so, if you want to multiple emotion in one audio, generate it separately for each emotion the merge all chunks. your reference audio should be around 3 seconds and need to carry only one emotion.

[@Jarods_Journey]

Awesome, thanks for the recommendation!

[@StefanReich]

Ah, these installation troubles. The hardest thing in computing today: Making pieces of software find the files they need. (I've tried mbrola etc too...)

[@Jarods_Journey]

90% of my time especially when installing things from GitHub 😅

[@SyntheticVoices]

Can't wait for fine-tuning. I've heard ChatGPT can generate training code if you provide the paper and code details.

[@Jarods_Journey]

Mmph... It's a quite a bit more involved than that. You still have to architect it, it will fail to generate code that completely works without you knowing what to tweak

[@4.0.4]

You've been my recent source of TTS news! Really neglected is voice recognition. Did anything dethrone Whisper for Japanese?

[@fulldivemedia]

Thanks again for the great work, love your channel

[@JustRayen]

16:53 Bro sounded like he manifested the voices in his head for a second XD

[@Jarods_Journey]

FR, or possibly, the model is trying to send us a subliminal message 😅

[@megaaziib]

the quality of the cloning is insanely good, but the vram usage is insane too. it using 15gb vram from my rtx 4060ti, the generation speed is quite fast 8-13 seconds.

[@Jarods_Journey]

Yeah, the VRAM usage is prohibitively large unfortunately, if it was 11, it might be more attractive

[@xbon1]

we need one click windows installer... but thank you for being first tutorial for how to do this on windows. I might actually try to do this now. edit: oh god this is way more complex than I thought with vscode edits, any chance you can publish the edited files & full tutorial somewhere?

[@dhillaz]

1:00 sounds like the narrator of youtube channel "Branch Education" (the one that does CGI teardowns of various modern technology)

[@flykiller]

Thanks again. If GPT-SoVITS weren't so undertrained (5k hours total for all 5 languages), it would probably be the best local model. It is great, but it struggles to pronounce difficult words or made-up ones like sci-fi alien words. Did you try training GPT-SoVITS further, as you did with F5?

[@Jarods_Journey]

I did do light finetuning of GPT-SoVITS and it's one of the better options out there for dialogue IMO. Not as much training as F5 though

[@jasonstetsonofficial]

Much better cloning of voice and audio quality

[@rawkeh]

Do you happen to know of any text-to-voice app/method that can change the voice? Like 50% Voice A, 30% Voice B, 20% Voice C? I want a voice that is not just a clone someone's voice

[@dustinsuburbia]

The style transfer could be useful if it is higher qual than RVC for generation emotion samples for F5

[@SyamsQbattar]

What is the difference between Aphion and F5-TTS?

[@AnasKhan-r8f8m]

I am still waiting for your tutorial on train gpt-sovits in new language.

[@speergz6452]

this might be a dumb question but i remember someone (may have been jarod) playing around with tts software that would default to a random voice or at least had that capability if not used for voice cloning. I'm trying to find a tts software that can do random voices rather than cloning but i havent found anything like that yet.

[@BeyondPC]

TortoiseTTS can do that.

[@chronicallychill9979]

How much time did you waste trying to get this whole thing installed? Could have just made a VM or container, no?

[@Jarods_Journey]

Hmm, probably like 2-3 hours just chasing packages around, mostly, just trying to get mbrola to work to no avail which is why I just skip it lol. It would've gone much smoother if I just installed it in wsl2 but I wanted to see what was wrong with windows

[@stonedape4538]

Looking forward to finetuning it in another language 😀 Hope it will be faster than f5.

[@JessicaSinclairDoomsday2024]

which free ai tts that sounds like conversation instead of reading a book tts?

[@py_man]

Notebook lm

[@jr-2nd]

Still sad that there are no free spanish TTS with that quality...

[@John-zy8ut]

can u please show us how its done with google collab or huggingfaces

[@Random_person_07]

This is awesome!

[@ItayHilel]

real time? finetuning new language?

[@GraveUypo]

so would you rank the offline options nowadays?

[@Jarods_Journey]

Will be ranking the Top 5 in December for local options

[@YannMetalhead]

Good video.

[@coc1841]

Not too bad!

[@BBZ101]

Is rtx 3060 enough for this mdoel ?

[@Jarods_Journey]

I don't think so due to maxing out at around 18 gb on inference when using it in the vid (22 was shown, but 4 was already being used so 18 just for MaskGCT)

[@BBZ101]

@Jarods_Journey thank you 😊

[@vixxcelacea2778]

Oh man, same here. I hope they make a version it so that it can still run. I don't care if it's slow to produce results. This is definitely one of the better ones I've heard.

[@jonogrimmer6013]

Change your duration

[@Jarods_Journey]

Hmm, probably