OpenAI Whisper - Robust Speech Recognition via Large-Scale Weak Supervision

Dive into a comprehensive video lecture exploring OpenAI's Whisper, a robust speech recognition system developed through large-scale weak supervision. Examine the paper's key findings and delve into the code implementation. Learn about the collection of a vast multi-lingual dataset, evaluation metrics, effective robustness, and scaling laws. Explore the model architecture, transcription tasks, mel spectrograms, language detection, and decoding heuristics. Gain insights into voice activity detection and token logit suppression techniques used in this cutting-edge automatic speech recognition system.

Intro
Paper overview
Collecting a large scale weakly supervised dataset
Evaluation metric issues WER
Effective robustness
Scaling laws in progress
Decoding is hacky
Code walk-through
Model architecture diagram vs code
Transcription task
Loading the audio, mel spectrograms
Language detection
Transcription task continued
Suppressing token logits
Voice activity detection
Decoding and heuristics
Outro