OpenAI Whisper - From Paper to Code

Following on from the previous week's unpacking of the Whisper paper (Robust Speech Recognition via Large-Scale Weak Supervision), in this session we switch gears and dive into the code.

🔧 What we’ll look at together:

• How the code reflects the encoder–decoder transformer design.
• Data preprocessing — handling noisy, imperfect transcripts.
• Training vs. inference setup.
• Practical adaptations — how to plug Whisper into real-world pipelines (e.g. multilingual transcription, noisy audio).

Please note that the session is not recorded.

Key links:

• Code
• Research Paper
• Zoom

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