MacWhisper has lengthy been probably the greatest and best methods to transcribe something on a Mac. I’ve relied on it closely because it launched, and I can’t suggest it sufficient. As we speak, it acquired even higher with an replace that provides help for Nvidia’s insanely quick Parakeet mannequin. And I imply quick.
As OpenAI strikes away from Whisper, MacWhisper will get a Parakeet enhance
When OpenAI got here out with its Whisper household of AI transcription fashions, they felt like magic. They had been a minimum of an order of magnitude sooner and extra correct than anything on the market, particularly within the open-source market.
Nonetheless, as OpenAI shifted its focus to the extra commercially viable gpt-4o-transcribe fashions, Whisper began to indicate its age. Quickly, newer transcription fashions did to Whisper what it had as soon as accomplished to its predecessors.
Nvidia’s Parakeet is considered one of them. Introduced simply final month, it promised to transcribe “60 minutes of audio in simply 1 second”, supplied you had the suitable {hardware}, corresponding to A100, H100, T4, or V100 GPUs.
Parakeet has been accessible by way of open-source instruments. The catch? Not everyone seems to be snug navigating command-line interfaces or managing customized mannequin deployments.
Fortunately, that modifications at the moment: MacWhisper simply added help for Nvidia’s Parakeet mannequin, and it truly is impossibly, extremely quick. Right here’s app developer Jordi Bruin:
“Due to our collaboration with the workforce at Argmax, MacWhisper now helps the Parakeet fashions. To indicate you what an enormous deal that is, try the gif under the place we transcribe and diarise a 30 minute podcast in beneath 8 seconds!”
I examined it on my M2 Professional MacBook Professional utilizing a latest 3-hour episode of 9to5Mac Blissful Hour, and it completed the job in simply 1 minute and 22 seconds, speaker recognition and all.
It’s price noting that the Parakeet mannequin is out there to Professional customers and at present helps English-only transcription. Bruin says the multilingual model is coming quickly.
FTC: We use earnings incomes auto affiliate hyperlinks. Extra.
