“Wie geht es dir?”
“I am good”
“Was ist dein Plan für heute?”
“I plan to read more about EzDubs!”
We’re pretty sure you put parts of this into Google Translate. But what if you could talk fluently, in real-time, to a German, in German?
That is the audacious goal of EzDubs.
We met Paddy, Amrut and Kareem, expecting them to have an asynchronous dubbing solution. As we tested the product, we uploaded one of our videos into EzDubs. We got our output in 10! 10 seconds, not minutes!
The latency of EzDubs was exceptionally low. In a market where latency is easily as high as 10 minutes, EzDubs had created “wow”. The founders’ deep experience building and researching in AI/ML for a collective 30 years showed. The trio had worked at Cisco, doing their masters in audio and speech machine learning from Columbia University.
The team saw a big opportunity in creating real-time translation. They identified a new underlying wave due to the creation of LLMs and exponentially increasing computational power. Real-time translation would now be possible, as easily as an app on your device. EzDubs was targeting a huge opportunity.
Translation services is a $45B market globally. Most translation services involve a human in the loop. The reason is accuracy and the need to preserve emotion, tone and rhythm. Known as prosody, the preservation of prosody through a machine is hard. Dubbing services charge as high as $500/hour, taking days due to this difficulty.
Imagine the unlock if you could do this in 1/5th the cost in a few seconds. EzDubs saw the power of low latency almost instantly.
EzDubs launched their tool for free via a Twitter bot to test their hypothesis. To say that it grew explosively is an understatement.The bot initially found pockets of fervent users like Indian football fans wanting to hear Messi’s world cup interviews in Hindi, and within a few months, it had grown to 100k followers, becoming one of the fastest growing AI bots on Twitter . Much larger, well-funded text-to-speech tools existed, yet people signed up for their dubbing service without a penny spent on marketing. Large enterprises found EzDubs thoroughly impressed with the output.
A flood of generative AI tools has emerged alongside EzDubs, doing audio, image, marketing or code. But many are essentially a UX layer built on top of a large language model. EzDubs, on the other hand, is building its audio model from scratch for real time sub-sentence translationa far more complex and ambitious approach. This will truly revolutionize how we communicate with each other in a multilingual environment.
Despite the love for their early product, the team believes they are just getting started. Paddy, Amrut, and Kareem’s conviction in an audacious goal, product thinking and technical horsepower made a compelling case for us.
Wir freuen uns über die Partnerschaft mit EzDubs.