Subscribe
Sign in
Unreal Speech
p/unreal-speechBetter and 8x Cheaper Text-to-Speech than AWS
Start new thread
Kevin William David

Unreal Speech - Better and 8x cheaper text-to-speech than AWS

by•

The first ultra-affordable, ultra-realistic AI narrator. Trained to sound audiobook-esque, perfect for narrating articles, blogs, newsletters, books, PDFs, and more! Compare against AWS side-by-side.

Add a comment

Replies

Best
Eric Jung
Maker
Hello everyone, This is Eric, the founder of Unreal Speech. Previously, I worked on Audioread (fka Audiblogs) which won #1 Product of the Day, Week, and Month and Golden Kitty 2021. While working on it, I found a huge problem. Hundreds of millions of people with reading difficulties rely on text-to-speech to read. Unfortunately, many are left with free but extremely low-quality text-to-speech. (e.g. On Mac, try highlighting any text, right-click, hit “Speech,” and “Start Speaking.”) 🥲 Higher-quality engines are practically monopolized by big tech companies who all magically agreed on the exact same exorbitant price point and use their massive resources to prevent competition. 😠 We are solving the 2 biggest challenges preventing the mass adoption of AI-based text-to-speech: 1) Too Expensive: ✅ solved 2) Too Robotic: ☑️ 80% solved (for non-fiction) We plan to 10x the dataset in the next 6 months. Not only will robotic voices be no longer a problem, you might even think it's a professional narrator. In the shorter term, we want to focus on helping TTS-enabled businesses succeed: read aloud apps (e.g. Pocket, Speechify, etc), UGC platforms (e.g. Medium, SubStack, etc), publications (e.g. Bloomberg, NY Times, etc), and e-learning platforms (e.g. Duolingo, Pearson, etc). We’re also interested in providing discounts to non-profits (e.g. Wikipedia, etc). 🤝 In the longer term, we want to make reading as easy as putting in your earbuds, for everyone, not just for those who have trouble reading. We believe that listening will be the new reading. And we’ll make reading as effortless as being a child whose parents read aloud to them. 💯 What Audible did to 0.2% of printed books, we are doing to the entire human and AI-generated corpus. Support us and help us fight back against the monopoly, provide an ultra-affordable, high-quality solution, and improve the lives of billions of people! 🙌 We would love to hear your feedback/questions/comments! Thanks, Eric
Faisal Albasu
@eric_j1 The Jordan Peterson and Gary Vee voices sound great, but one thing I'm curious about is the ruling on using voices of real people. Since this is essentially a deepfake, is there any contract or agreement between Unreal Speech and these guys on using their voices?
Anna Filou
@albas @eric_j1 I’m not sure if the “seek forgiveness later” approach would work in your favor in case of a lawsuit… The two voices (Entrepreneur and Professor) sound great, but I can imagine the owners of the voices getting really angry upon learning that you’re basically allowing anyone to deep-fake their voice. I know I would be.
Thomas Skavhellen
Wow, the Gary Vee (Entrtpenour) voice and Jordan Peterson (Professor) voices are unreal. I can basically get Gery Vee to read all my blog articles for this :) Super cool!
Eric Jung
Maker
@skavhellen it be cool to have him yell at you all the motivational articles especially
Erik Dunteman
The quality of these voices is absurdly good. I'm in the ML space myself and training models like this takes SOOOOO much good data. I'm impressed. Also your biz history with Audiblogs is legit. Cool to see you solving your own problems. Best of luck on the launch!
Eric Jung
Maker
@erikdoingthings Thanks for the feedback Erik!
Vitaly Matveev
Very nice. 12.5m number seems more like for publishers than for end users. As an end user I'd buy a fraction of it at circa 9 bucks per month.
Eric Jung
Maker
@vitaly2016 Yeah, totally! At the moment, we're targeting companies who'd use our tech to provide services to end-users for different use cases!
Kyle Morris
Wow! You've surpassed the Siri/AWS deadpan "AI assistant" voice & made something that sounds like a recording + playback of an actual person Just to clarify, this is AI generated right?
Kyle Morris
Also listening to the Professor voice read out the Bee Movie script and it feels like a lecture in evolutionary biology
Eric Jung
Maker
@morriscode lol haha it is
Eric Jung
Maker
@morriscode this is precisely the use case I envisioned.
JP
Is that Jordan Peterson and Gary Vee that I hear? 😂
Eric Jung
Maker
@dynamo 🤭
Tom P.
Fantastic voices. However I would humbly suggest to add a pay per character/word plan and an interface to simply copy/paste text. Or contact a kind of reseller (voicemaker has the interface, they would just need to buy your plan in addition to the other voices they buy. PS I suppose only English language so far?
Eric Jung
Maker
@torsten_p I appreciate the humble suggestion! A pay-as-you-go option is definitely a possibility in the future, maybe as early as in the next month or two. But yeah, we're focusing on working with "resellers" and helping them create various use cases like that. Currently, we have only English. Our model is multi-lingual, though, and we plan to add more languages soon.
Anson Kao
Loving the voices, they sound wayyy more familiar and pleasant than most options I've heard! Way to go!
Eric Jung
Maker
@anson_kao Thanks for checking it out! We definitely tried to be more familiar and pleasant.
Vasilii
The sound quality is amazing! It sounds almost like human speech. I wonder if it's really possible to make a solution that will work without the Internet.
Eric Jung
Maker
@milovidov_vasya That'll certainly happen, but such a model will always be lower in quality than models that run on GPUs. So I think at some point, pretty high-quality models will be available on your phone or something, but you'll also have access to even higher quality models that are probably only available over the Internet!
Marin Adendall
Cringey choices for celebrity typecasts. Cool Project
Eric Jung
Maker
@peppereddirt At least we tried 😅 Thanks for the comment!
Foma Kinaev
Cheaper than aws - ok, but what about performance and what kind of AI is on the backend?
Eric Jung
Maker
@owlycrap That's a fantastic question, and yeah, performance-wise, AWS has more scalability. We're going to focus and partner with a small number of companies and iterate quickly to make sure we perform at a level they expect. We did partner with a serverless ML hosting company to be able to scale fairly quickly.
Muskan Thakur
@eric_j1 Heyy This is such a great Application! Like Human Speech, pleasebring more of such creative content! Congratulations on your launch!
Eric Jung
Maker
@muskan_thakur Thank you so much! It'll only become more human over time.
Amelia Charlie
Congratulations on the launch team Unreal Speech
Eric Jung
Maker
@amelia_charlie thank you for the support.
Basharath
This is so good. The sounds appear so realistic. Congrats on the launch.
Eric Jung
Maker
@basharath thank you so much! And it'll only get more realistic.
Devluc
Had fun with it hearing how potential domain names for my projects would sound like in context :) Sounds so natural. Congratulation for your work and good luck with the launch
Eric Jung
Maker
@luciantartea haha, cool! Not sure if it can pronounce domain names that well... But hopefully, it helped.
Hélène SAN
Hey, Congrats on your launch! 🚀 We will be launching soon too, (@snackeet ), I hope that you will support us as well, 💛
Eric Jung
Maker
(@snackeet @helene_san will do! Seems super interesting.
Kurt Bonatz
Great demo site. Do you have ability to tag words inbound/outbound for emphasis. For instance, if you play the sentence: "Prime Minister Theresa May worked diligently on her new legislative agenda" it cannot understand that Theresa May is a person and as such does not articulate it correctly thereby missing the context / meaning.
Eric Jung
Maker
@bgr4261 This is a fantastic question. We currently don't, but we'll definitely be able to support SSML to enable something like this.
Kurt Bonatz
@eric_j1 Great to hear! Let's talk - our AI (Applied General Intelligence) understands meaning and deep semantics and we are looking for a V2T, T2V solution to demo our deep understanding abilities.
Eric Jung
Maker
@bgr4261 what are T2V / T2V?
Florian Hidayat
This is really awesome, Eric. Great work! I was playing around with it and found something you can improve upon though. Try inputting "Rafael Nadal Parera is a Spanish professional tennis player. He is ranked world No. 4 in singles by the Association of Tennis Professionals; he has previously been ranked world No. 1 for 209 weeks and finished as the year-end No. 1 five times." 😉
Eric Jung
Maker
@florian_pranata_hidayat Got it, thanks for reporting this! This is a rather easy fix 😉
Andrew Glenn
Took me a few efforts to listen to the samples, but I finally got them. Also, I've read the thread over on ycombinator. A few thoughts: The professor and entrepreneur (obviously modeled after Jordan Peterson and Gary Vaynerchuk, respectively) sound excellent. The others all sounded fine too. I didn't have the artifacts that the others complained off on ycombinator. Even if there are artifacts or otherwise imperfect audio, It still represents a great MVP that you can go to market with while continuing to innovate and refine. Kudos @eric_j1!
Eric Jung
Maker
@asglenn 🤙🤙🤙yay, thank you so much for the feedback, this is probably one of the best comments I've got! I think calling it "better" really triggered people and made them look for defects, as opposed to if I just pitched the "8x cheaper," maybe people could've been more pleasantly surprised by how good the voice was. I do hope this is a good enough MVP (probably the least minimal MVP I've built so far) to iterate on. Thank you so much for checking it out and taking the time to write the comment!
Daniel
Can you also make a French voice as well so interested to use it.