OpenAI Now Has a Text-to-Speech API - Testing and Overview

[@megaaziib]

the voice is good, but it's not good as elevenlabs level, elevenlabs voice can have different intonation according to text and punctuation (very good for roleplay chatbot). open ai voice is similar to micosoft edge tts which you can use it for free.

[@KeystoneScience]

So excited for this. I hope they come out with a nice zeroshot cloning model to use that is at a similar price point.

[@Jarods_Journey]

This would be pretty cool! Dunno if they're thinking about any liability for that though lol

[@Mowgi]

That chat assistant in your own example sure did use a lot more words than necessary :P Your face at 7:55 🤣 Looks exciting, keen to see where it goes from here. As always, interested to see if they make any advances with accents.

[@Jarods_Journey]

Haha ofc. Accents will hopefully be coming, that'll be a nice addition!

[@shawn4990]

First time catching your content Jarod... great Video. Easy subscribe. In reading the comments, I noticed many users are finding the limitations of OpenAi's software. Like anything else, early adopters/users of new technology share in the often painful growth of such tech while its development matures and evolves. For many though, they see this as a 'glass' half empty, rather than half full... either way, in my humble opinion, our future looks bright.

[@Jarods_Journey]

I agree, I'm generally optimistic on these things. Looking at the same time last year or the year before, the "AI" growth has been awesome, so it's just developing and maturing

[@myte1why]

well for Turkish I can say it sound like non native speaker talking but not bad. to explain it: In native Turkish speakers, words tend to flow more smoothly, whereas non-native speakers often have a tendency to place more emphasis on words. The situation here is the same. That is to say, it's quite good but for someone whose native language is different.

[@stevecato]

Not really clear where OpenAI want to go with this being so late to the party. I've been using EdgeTTS, Piper and looking forward to your 11Labs replacement - can't really see why I would pay OpenAI to use this one.

[@Jarods_Journey]

Always room for competition and so I think it's about time they get in the field. Quality wise it's not bad as it's still relatively lifelike for many languages, but I wouldn't use it for personal stuff as I've got tortoise+rvc. BuT the creator of tortoise is part of OpenAI sooo I can see things getting pretty good over this next year

[@doingtime20]

I tried on spanish, my native language, and honestly it's pretty good. Only issue is all voices speak like an american with a USA english accent. The better is one is Nova, very little accent.

[@Urfriendinfinance]

I saw your tortoise TTS video. What's the best AI voice right now is it? heard to use Tortoise TTS then rvc voice.

[@Jarods_Journey]

I haven't heard anything really much better than Tortoise TTS to RVC. XTTS v2 is also good. Eleven labs if you don't mind paying.

[@Urfriendinfinance]

@@Jarods_Journey thanks for the quick reply. i was reading your about page and noticed you graduated from chico. I also graduated from there 2010.

[@stan-zm3ep]

thank you Jarrod for all your contribution, u deserve a PhD in TTS :)

[@Jarods_Journey]

Haha thanks!

[@ИванКоряко-ц5б]

And what about streaming? Could you get audio responses until full text generation?

[@piersonmarks]

Yes - streaming is available via chunk transfer encoding. You can just stream the response from OpenAI to a buffer/file

[@Jordan-tr3fn]

cool video ! how did you write in japanese like that at 7:38 ? I'm learning japanese but find it difficult to write without another virtual keyboard

[@Jarods_Journey]

You can install the microsoft keyboard input and then when you type, you just use romaji to type. Search around online to see how you can get it up and running :)!

[@Jordan-tr3fn]

@@Jarods_Journey thank you!

[@nufh]

The Indonesia is quite spot on.

[@wisnunugroho9958]

berbayar ga bang ini? Kalo dijalanin lokal gini..

[@juancalara3997]

Interesting! is there a way to add this to a custom GPT?

[@kaeuzita]

The AI speaks other languages like an American trying to speak other languages. The accent lol

[@rickarroyo]

09:29 - In Portuguese it doesn't sound like Brazil or Portugal, it sounds strange 😑 But your like is guaranteed :))

[@MaisnerProductions]

Great update. Interesting.

[@Jarods_Journey]

Thanks mate :)!

[@andreimariano2674]

I am a native Filipino speaker and I would say Open AI always know how to play their cards even with multilingual tts. Really excited for this.

[@naturalbeauty19964]

How can I add any voice , for example voice from marvel iron man and any persons ?

[@Jarods_Journey]

Not possible, I don't think OpenAI wants to be liable for voice cloning lol

[@mythaimusic39]

I played with the API yesterday and let me tell you that I have a nearly perfect french with it. I tried every voices and they all are nearly perfect with just a (very) slight English accent here and there. That's far from your demo where the french speaking is horrible, so I guess that's because it is speaking several languages in the same prompt ?!?

[@Jarods_Journey]

Yep! You're on the money there, I noticed the same thing afterwards where when you use one language, it's much better. Probably a transformer model and is inferring an English accent or a mixture of accents when I used it in the example

[@tacototo8933]

Dutch was all right, but pronounced "eten" (eat) incorrectly. Nontheless this is amazing

[@bobbyboe]

german = bad french = bad italian = bad the voice in your app = very good 😊 I wonder if you enable your audiobook-app to speak german... and make it installable for non-coders like me 🙏 I would like to train my own voice and be able to do recordings without setting up my audio-setup.

[@YumiSumire]

Vietnamese is really accurate

[@SpudHead42]

Android only so far for the chatGPT tts?

[@Jarods_Journey]

I actually am not sure, I'd assume it's availble on iOS as well

[@Airbender131090]

Can we run it localy?

[@Jarods_Journey]

Yes, you just need to call the API

[@jerrythefeared]

French, German and Dutch were horrendous. Korean started fine, but the other 3/4th was butchered really bad :/ edit: Russian was good!

[@mikeiavelli]

French 9:06 ; 10:18 --> Sadly, the french version was really bad. Maybe it's better with another voice?

[@danielmonge2318]

I'm from Brazil and let me just say... That Portuguese was terrible lol. It had a heavy American accent. I've played around and I also like Nova the best. When done in isolation, with a text input that is 100% Portuguese, it does a lot better, but it does have a country-side accent to it, but at least it does sound like a native. I don't like the fact that it is too fast. You can control the speed by sending the "speed" and a float, from 0.25 to 4.0, where 1.0 is normal speed, but that's just an audio post-processing step, like changing the speed on a KZbin video.

[@Jarods_Journey]

Yeah, I noticed too that if you use english and then have it switch to another language, it might start speaking english with a tinge of an accent or vise versa. They're probably running a transformer model behind their TTS as well like tortoise so this could be affecting the output on it haha.

[@WanJieMing]

You don't eat dairy ?? WHY SO WEAK ?

[@bluevisor]

What kind of Asian uses calculator for 1.5x30?

[@gamersgabangest3179]

I am italian and the Italian accent is not perfect

[@m_a_p]

Most of the non-english TTS sounds like an American trying more or less hard to speak the language. German and French are especially terrible.

[@MarcoManzo]

in german it has a strong english accent. not great

[@Airbender131090]

Russian is very bad… crazy english accent