ChatTTS - Best Quality Open Source Text-to-Speech Model? | Tutorial + Ollama Setup

ChatTTS - Best Quality Open Source Text-to-Speech Model? | Tutorial + Ollama Setup
👊 Become a member and get access to GitHub and Code:
/ allaboutai
🤖 Great AI Engineer Course:
scrimba.com/learn/aiengineer?...
🔥 Open GitHub Repos:
github.com/AllAboutAI-YT/easy...
📧 Join the newsletter:
www.allabtai.com/newsletter/
🌐 My website:
www.allabtai.com
Code in Video:
github.com/AllAboutAI-YT/chatTTS
ChatTTS GitHub: github.com/2noise/ChatTTS
Tutorial on how to install and use the Open Source TTS model ChatTTS. Very fun to play around with and a great addition to the OS TTS landscape.
00:00 ChatTTS Intro
00:57 Install / Setup ChatTTS
03:30 Testing
06:28 ChatTTS with Ollama (Llama 3)
08:58 Conclusion

[@Ms.Robot.]

Totally nailed this tut.This was very well explained. It was organized in order, easy to understand step-by-step instructions, and it addressed important points in case we run into obstacles.❤ (Still waiting for gpt4o voice too. I hope customizable voices are an option.)

[@AllAboutAI]

thansk a lot :D really appericiate it

[@john5s]

I found that by setting a seed I can keep the voice sounding the same. chat = ChatTTS.Chat() chat.load_models(compile=True) # Set to True for better performance torch.manual_seed(seedNumber)

[@2099EK]

You're a hero! I wish more people would check out this comment. Thank you for posting this!

[@neolynxer]

Great stuff. Please implement a "a... yeah" counter in your videos! :D should be fun.

[@sophiedelavelle5958]

26 only according to the transcript

[@DaCashRap]

Uh yeah, great video overall!

[@EricB1]

Great find. One suggestion: remove all [yeah] from your script.

[@joachimschoder]

The project libukai/Awesome-ChatTTS has a more extensive documentation. It is in Chinese but Google Chrome can automatically translate at least the text elements. It doesn't replace a good english documentation but it is a good starting point.

[@darcwader]

very nice

[@ArseniyPotapov]

XTTSv2 is also very good model, highly recommended

[@korni5149]

How can I achieve this gradient text color and text appearing animation in Windows Terminal?

[@micbab-vg2mu]

the quality is great - I have to try it

[@AllAboutAI]

yes give it a go :)

[@legendarystuff6971]

By default it generates a voice randomly from gaussian noise, you can definitely chose the voice somehow, look over the example they have on their repo and ask Opus for help. Their materials combine English and Chinese which makes it a bit annoying. On they're bilibili video, whatever bilibili is, they even clone Steve job's voice and Taylor swift. Nice find, thank you

[@2099EK]

You'd've been a hero if you just put that here.

[@user-um2uq9nh4z]

Hi! If english is still experimental then french or german...

[@mendthedivide]

works well, but only with smaller wording/sentences. almost sounds real at times, nice find!

[@AllAboutAI]

thnx :) yeah i think the token limit is like 380

[@elgodric]

Is there a webui or something for non coders

[@NirmalEleQtra]

Can we have an Indian English accent or any other Indian language dialect here? If yes, how can we do it?

[@limebulls]

Can it speak german etc as well?

[@mendthedivide]

It supports both Chinese and English

[@Tofu3435]

I want to make an audiobook from light novels. My phones built in reader are too robotic, natural reader are needs internet access and too expensive. Maybe a program built on this can help.

[@MLDQ]

Is it better than realtimeTTS XTTS with coqui?

[@AllAboutAI]

like i said in the vid, i dont think I could use this for real time s t s because of the compute time

[@mrpro7737]

very nice base voices i can voice chnge them to any model i want

[@_zproxy]

how

[@AllAboutAI]

do you know how?

[@mrpro7737]

@@AllAboutAI i am using this open source project h§ttps§://youtu.§be/nXpBlC6OBw4?si=UOQ86u97CLp0BgJN for training models and voice change because its only support audio to audio and applio§.org for text to speech because it support so many accents and languages remove the § from urls , i post this comment 3 times and yt keep deleting it 😑

[@mrpro7737]

@AllAboutAI i am using this open source project h§ttps§://youtu.§be/nXpBlC6OBw4?si=UOQ86u97CLp0BgJN for training models and voice change because its only support audio to audio and applio§.org for text to speech because it support so many accents and languages remove the § from urls , i post this comment 3 times and yt keep deleting it 😑

[@Edward_ZS]

Did anyone find a way to,select the voice?

[@snatvb]

similar to "bark" from suno interesting which is better and perfomance

[@AllAboutAI]

noted, tnx :)

[@christiandarkin]

bark's problem - when i tested it at least - was that it didn't always say what you told it to say. often it just made stuff up.

[@snatvb]

@@christiandarkin I faced with this if I passed to long text

[@OliNorwell]

It's interesting how the best quality settings aren't used in the basic advanced demo... I thought those later samples were very good. Not sure how it got so many stars so quick on Github though, I mean, there are alternatives that are excellent that didn't grow so quick at all

[@zachary3603]

What are the better alternatives everyone is using? Using Elevenlabs atm, but the speech comes out so robotic half the time.

[@luisvictorf]

cool story about the underwater cats, would've liked to hear a bit more; it would make a good kids story! =D

[@AllAboutAI]

haha yes

[@glikoz]

How about speech-to-text ?

[@VaibhavShewale]

cant set the specific speaker

[@AllAboutAI]

yeah thats the issue right

[@Project_SaveTheWorld]

It be nice if you could have it read Spanish. If it could, you'd pretty much have a translator.

[@AEnoob]

i think there is a Text Seed that lets you choose your voice

[@AllAboutAI]

oh nice, will look for it

[@nexuslux]

What’s the best real time equivalent of this?

[@Resursator]

GPT 4o, probably.

[@BORCHLEO]

coqui/tts with the xtts model

[@NathanChambers]

I've been really happy with alltalk_tts with deepspeed enabled. It's api worked good for my personal needs too. I'm no pro with tts stuff but it's been great for me. I have it reading/speaking in English, German, and Russian and does great. The only issues I've ever really had with it is if it's set in "ru" but reading "en" text.... punctuation can become demonic sounds :P It's Russian accent speaking English words is a real bonus to me too. :P

[@AllAboutAI]

yeah xtts is great

[@agatitytube]

I assume, it is available only in english, right?

[@mendthedivide]

It supports both Chinese and English

[@beliebigerusername]

how can i train it a different language? or connect it to another model?

[@DihelsonMendonca]

No Brazilian Portuguese, unfortunately 😮😮

[@AllAboutAI]

yeah only en and ch for now i think

[@8eck]

Don't see any weights on huggingface, so don't understand all the hype... But the code looks real.

[@stanTrX]

Does it speak Turkish?

[@AllAboutAI]

i think its only english and chineese atm

[@kamalkamals]

but not support most popular languages

[@dumbol8126]

chinese ml engineers are goated

[@gustavheinrich5565]

Just make sure you're not falling for llms with Chinese propaganda and false information baked in, when using that stuff.

[@threepe0]

Turtle is better I think

[@Ginto_O]

Not good

[@Sanguen666]

tortoise is still better

[@kobe81]

ChatTTS.model.gpt:Incomplete result. hit max_new_token: 384 too bad...

[@JNET_Reloaded]

nice can you add the requirements.txt file ChatTTS IPython requests openai torchaudio torch numpy omegaconf vocos vector_quantize_pytorch transformers

[@AllAboutAI]

ill try to do that :)

[@OUTLANDAH]

@@AllAboutAI im having the same issue when i get to adding the requirments i get an error.