Show HN: Built speech models for Europe, accidentally topped Open-ASR in English

Category: ai-ml

Tags: speech-recognition, machine-learning, inference-optimization

Score: 8.0/10 (Innovation: 7, Technical: 9, Documentation: 4, Utility: 8)

Resonant-1 and Resonant-1-flash are state-of-the-art speech recognition models that top the Open ASR Leaderboard for English and excel across multiple European languages. The project is interesting because it combines novel reinforcement learning training techniques with deep inference optimizations through a custom engine called Violin, achieving remarkable speed (processing one hour of audio in under 3 seconds) while maintaining leading accuracy.

Target audience: data engineers, ai researchers, backend devs

Repository: https://www.reson8.dev/blog/benchmarks

View on Hacker News