Whisper by OpenAI is an open-source speech recognition system trained on thousands of hours of data. It provides English subtitles with 95%+ accuracy, for videos in any language in real-time. Dubverse is bringing this technology to you as a product!
Hello Hunters, Makers, and product Geeks 👋
Anuja here, co-founder of Dubverse.ai
I am super excited to announce our new subtitling tool, powered by OpenAI’s Whisper!
Whisper is an open-source speech recognition system trained on thousands of hours of data, and it’s able to provide accurate English subtitles for videos in any language.
For those new to Dubverse, a quick intro - we are a team of ex-founders who believe in the power of connection through languages. We are a deep-tech SaaS with a focus on building translation and VoiceOver engines. One of our products is the Dubverse Webapp which takes in a video and instantly dubs it into other languages at scale.
We’re proud to be able to offer this innovative tool to our users.
Let us know your thoughts in the comments!
Report
@anuja_dhawan2 This is amazing! Do you also have the feature to burn the subtitles in the video? We do a lot of Spanish/French Shorts type content. It would be a great add.
Report
@anuja_dhawan2 Sounds good, results are preety good I tried it for some of my videos
Report
@anuja_dhawan2 being an early adopter I can't stop raving about the product and the company. All the best
Hi Hunters! I’m Varshul, Cofounder of Dubverse 🙋♂️
Whisper was open sourced on 21st Sept. We always keep talking about “breaking the language barrier,” and it was a no-brainer for us to jump onto this right away.
Our experiments with different types of video files gave outputs that required almost negligible editing. The Dubverse Webapp already had capability of generating translated subtitles, but it required heavy editing as proper nouns were not identified correctly. We hacked Whisper into our product by pulling an all-nighter the very same day.
After a day’s internal testing, we invited some of our users to try it out. To our [excitement?], we saw them subtitling ~30 videos a day for their YouTube channels. As of the launch, 15 alpha users have subbed ~20+ hours of YouTube videos in English.
The most common feedback : THIS IS DAMN ACCURATE! 🤯
Adding on @anuja_dhawan2 's comment above, I wanted to share with you the some major features you’ll enjoy:
✨ Project Management — organize your subs into folders and share them with your team
🎥 YouTube Picture-in-Picture — play the video side by side to make the reviews easier
📝 Transcripts Editor — Edit the English output, down to the timestamps with basic utilities like find and replace
You’ll discover tone of other features and improvements that will make your content creation journey super easy. 🌟
What do you think we can improve upon it? Got additional feature requests? Comment below!
Hello, fellow tech geeks! 🙋♂️ Myself Jaskaran and I am an AI engineer at Dubverse. Here to highlight what fascinates me about the OpenAI Whisper tech
Whisper is trained on 680,000 hours 🤯 of labeled audio, which makes the dataset one of the largest ever created in supervised speech recognition.
In addition to that, the model is also trained on the translation task that is [x language] -> English (125,739 hours of data).
The amount of huge data whisper is trained on allows it to be accurate in transcription and translation part directly from speech.
Had we used a transcription + google translate pipeline, the error would propagate which would need a human in the loop.
Whisper solves it in a one-shot style that is directly from speech and thus does not have cascaded errors. It is also amazingly good at transcribing/translating Songs 🤯.
Excited to see the mass adoption of the tech.
Whispering Dubverse, Powered by OpenAI
Whispering Dubverse, Powered by OpenAI
Donate to India's Covid Relief
Whispering Dubverse, Powered by OpenAI
Whispering Dubverse, Powered by OpenAI
Whispering Dubverse, Powered by OpenAI
Whispering Dubverse, Powered by OpenAI
Whispering Dubverse, Powered by OpenAI
Whispering Dubverse, Powered by OpenAI