Research Paper review (OpenAI Whisper)

Join us to review OpenAI's groundbreaking paper that revolutionised automatic speech recognition and underpins the Whisper speech-to-text model:

• Robust Speech Recognition via Large-Scale Weak Supervision

We'll aim to dissect the methodology, architecture, and implications of training on multilingual audio data.

Key Points:

• Weak supervision approach - How Whisper learned from imperfect, web-scraped data
• Zero-shot transfer capabilities - Why it works across languages without fine-tuning
• Architecture deep dive - Transformer encoder-decoder design choices
• Scaling insights - What 680k hours of data teaches us about speech AI
• Real-world performance - Robustness to accents, noise, and domain shifts

Please note that the session is not recorded.

Key links:

• Research Paper: Robust Speech Recognition via Large-Scale Weak Supervision

Discord joining instructions: https://bit.ly/llm-discord