Gemini 4 Argon → https://g.dev/ai/gemini-about
Google Gemini models → https://g.dev/ai/gemini-models-docs
Watch the full episode → https://g.dev/youtube/gemini-models-full
In this video, Luke Schlangen and Mollie Pettit team up with fellow Googler Gabe Weiss on a classically difficult problem in audio machine learning: transcribing a multi-instrument, polyphonic audio track into playable sheet music.
Watch as they throw it at three different Gemini models (Gemini 3.5 Flash-Lite, Gemini 3.8 Flash, and the new Gemini 4 Argon) and listen to Gabe play the results on the mandolin.
Chapters:
0:00 – Gabe’s challenge: Transcribing tunes from audio to mandolin sheet music
1:30 – Gemini Flash-Lite: Wrong key, wrong clefs, and definitely not the song
2:10 – Gemini Flash: Right key, closer structure, but reversed sections
3:19 – Gemini 4 (Argon): It's pretty close.
4:18 – Hearing the actual tune: How close did Gemini 4 really get?
5:20 – Cracking the "Polyphonic Problem": Why this is classically hard for AI
More resources:
Learn more about Gemini 3.8 Flash → https://goo.gle/3TuVfmT
Builders → https://g.dev/ai/builder-member-awards
Watch more The Agent Factory → https://www.youtube.com/playlist?list=PLIivdWyY5sqLXR1eSkiM5bE6pFlXC-OSs
? Subscribe to Google Cloud Tech → https://goo.gle/GoogleCloudTech
#GoogleCloud #Gemini #GeminiAtWork #Gemini4Argon #GeminiFlash #AIAgents #GenerativeAI
Speakers: Luke Schlangen, Mollie Pettit, Gabe Weiss
Products Mentioned: Gemini, Gemini 4 Argon, Gemini 3.8 Flash, Gemini 3.5 Flash-Lite, Google Antigravity