Whisper is an open source AI model for multi-language speech recognition developed by OpenAI and trained on 680,000 hours of audio in almost 100 different languages. With moderate computational requirements, it achieves near state-of-the-art performance compared to both other open source models and commercial systems.