Can Gemini transcribe a song?

Why watch

Google engineers test three Gemini models (3.5 Flash-Lite, 3.8 Flash, and 4 Argon) on transcribing a mandolin audio track into playable sheet music. The video demonstrates incremental improvement across models, with Gemini 4 Argon getting closest to the original tune, while explaining why polyphonic music transcription remains a challenging problem in audio machine learning. Developers interested in audio ML and generative AI capabilities will learn about current model limitations and progress.