17/02/2026
We open-sourced two EN↔JA speech translation models on Hugging Face:
🎯 Qwen3-ASR (1.7B) — 4.2/5 quality ⚡ Distilled Whisper (756M) — 4.6x faster
Benchmarked vs Whisper large-v3 & SeamlessM4T v2.
Benchmarking bidirectional English-Japanese speech translation models — Qwen3-ASR (1.7B, highest quality) vs distilled Whisper (756M, 4x faster) — against OpenAI Whisper large-v3 and Meta SeamlessM4T v2