How OpenAI trained a speech recognition model on 680,000 hours of audio data.

OpenAI’s Whisper Learned 680,000 Hours Of Speech! · 9:00 · Video: Two Minute Papers · Watch at the source

In short: This video examines the technical approach behind OpenAI's Whisper, a system designed for high-performance speech recognition. By leveraging a massive, diverse dataset, the model achieves impressive results in transcribing spoken language, demonstrating the power of large-scale weak supervision in modern machine learning.

The research, titled 'Robust Speech Recognition via Large-Scale Weak Supervision,' details how the model was trained on an extensive collection of 680,000 hours of audio. This vast dataset allows the system to generalize across various accents, background noises, and technical jargon, making it highly versatile for real-world applications.

By utilizing weak supervision, the developers bypassed the need for perfectly curated, human-labeled transcripts. Instead, the model learned from the sheer volume of web-based audio and its corresponding text, showcasing a significant shift in how we approach building robust artificial intelligence for language processing.

Opens the Intellect+ feed with this card first. Swipe up for the next one.

Sources

To cite this card, link to https://intellectplus.co.uk/app/technology-openai-s-whisper-learned-680-000-hours-of-speech.

How we source, check and correct cards: Editorial standards.

General information, not professional advice. Spotted an error? Flag the card or see corrections and complaints.

Intellect Plus is a visual educational discovery platform for exploring fascinating facts and ideas. Brain Food, Not Brain Rot. What is Intellect Plus?

Related

Finding something worth knowing…