Groq reposted this
Government Solution Architect | E government consultant | Emerging Technology Advisor | Technology Analyst
Just tried out Groq Speech-to-Text, and I am thoroughly impressed! I fed it a 15-minute Arabic podcast, and it transcribed the entire thing in just 5 seconds. The speed and accuracy are truly remarkable. Groq, you never fail to amaze with your cutting-edge technology and innovative solutions. For context, the Groq Whisper Large V3 operates at an astounding ~172x speed factor and costs only $0.03 per hour transcribed. In comparison, OpenAI's service costs $0.006 per minute and $0.36 per hour, and is significantly slower than Groq. This efficiency and cost-effectiveness make Groq's solution a game-changer for anyone working with audio content. #SpeechToText #AI #MachineLearning #Groq #Productivity #Innovation #TechExcellence #CostEfficiency
They have been a game changer in inference speed, and having open source models hosted in their cloud is such a smart move, focusing on the hardware more than what runs on it.
I wrote this article in the same context "Unlocking Open Source Voices: Evaluation of Whisper, wav2vec 2.0, and Kaldi ASR Models", hopefully it will give bird view https://www.linkedin.com/pulse/unlocking-open-source-voices-evaluation-whisper-wav2vec-ben-nasr-owwqf?utm_source=share&utm_medium=member_android&utm_campaign=share_via
That's 1,100% more expensive than using Groq - those are crazy numbers. The question is there is an incremental decrease from a certain volume or not, because Whisper is OpenAI's. Commercialising but then another company sells it for much cheaper is a completely different game altogether.
Vuong Ngo
1w
amazing performance. Wondering if you benchmarked streaming as well? Would be super useful for many cases
HariHaraSudhan KM has been yapping non stop about Groq . Cant get him to stop,
Oskar Mencer
1w
Groq game changer!
Literally "insane". 😇
Uven Moodley
1w
Useful tips
Technology Executive | Fullstack Software Architect | AI Enthusiast
1wAny known solution to make the models tag the speakers ?