Robots will take over the world…one day at least! If you don’t believe it by now, you should consider that thought. It starts small and then BAM! It takes over, just like ChatGPT did and now Whisper.
Whisper is the new hip AI tool that will be added to the never-ending list of AI software that both make our lives easier and might steal our jobs. But let’s look at the bright side for now.
Whisper is “an automatic speech recognition (ASR) system trained on 680,000 hours of multilingual and multitask supervised data collected from the web.” It is brought to you by – the one and only- OpenAI. Yes, the same company that brought you ChatGPT and threatening most AI startups at the moment.
The Florence Pugh of ASR
It’s basically speech-to-text software, but what makes Whisper so special?! I mean, even Android users have that on their phones (no shade, though!). Well, to begin with, it’s an encoder-decoder Transformer, the same tech that was used in creating ChatGPT. You can read all about it here!
Whisper was trained for 680,000 hours!! That’s 77.6 years, I’m gonna take a chance a say that this younger than both of us. Unlike other speech-to-text software, it also doesn’t believe in English language supremacy! Yes, it does work with other existing languages!!
It can also translate from several languages into English. Arabic, Tamil, Spanish, Thai, Lithuanian, Kazakh, Swahili, Galician, have at it! Just like Florence Pugh, it is remarkable!
Unlike ChatGPT, the mothership released Whisper as its first open-source model. That means anyone can tweak and modify the code however they see fit. This is important because (in simple words) OpenAI create insane deep learning models but they’re often not open-sourced. So this step with Whisper can allow for more research on deep learning models and could lead to the creation of specialized applications.
It's impressive as it has become the norm of AI and if you want to test it, knock yourself out!