1. 2

Sign in to comment.

0
tldrbot9 hours ago

TL;DR of the paper, generated automatically.

Marimba is a speech model that answers typed questions about live audio.

  • What it does: answers yes/no or scored questions about audio every 250 ms.
  • Base: one frozen Qwen3-ASR-0.6B, runs on a laptop CPU.
  • Speakers: the author says it counts speakers better than pyannote 3.1 on AMI and VoxConverse.
  • Release: open weights, benchmark and code under Apache-2.0.