Rumble-2-Flash: open-weights speech-to-text model handles accents, noise and 40+ languages
omarsar0 · x · 2026-11-18
Reson8 released Rumble-2-Flash, a free open-weights speech-to-text model trained on messy real-world audio — heavy accents, crosstalk, phone calls, background noise.
- Covers 40+ languages
- Apache 2.0 weights, runs locally so audio stays on-device
- The creator claims it beats models 20x its size and is small enough to run on a phone
More from Multimodal
- Runway cuts real-time 57B video model serving from 8 H100s to 4 with inference optimizations — runwayml · 2026-11-18
- Open-weights ASR model Rumble-2-Flash hits #9 on Hugging Face Open ASR leaderboard, 40+ languages — cneuralnetwork · 2026-11-18
- Gemini 3.5 Transcribe adds real-time voice-command editing for dictation — AI_Andrew · 2026-11-18
- Antipode: An Open-Source Local Synth Controlled via MCP, Turning Agent Ideas into WAVs — Dizzy_Basil6892 · 2026-11-18
- Audio EQ and compression inside Antigravity: Gemini 4 builds in-chat FFmpeg console — alexanderchen · 2026-11-18
- Midjourney 5-Style sref Mix Formula Yields Cinematic Film-Look Batmobile Shots — michaelrabone · 2026-11-18