Introducing the FFASR Leaderboard: Benchmarking ASR in the Real World

The Far-Field ASR (FFASR) Leaderboard is a new community-driven benchmark for evaluating Automatic Speech Recognition models under realistic far-field acoustic conditions. It aims to bridge the gap between standard clean-speech evaluations and real-world performance by incorporating factors like reverberation, background noise, and microphone distance. The benchmark utilizes a hybrid simulation engine to create 14 distinct acoustic scenes with varying signal-to-noise ratios. Submissions are evaluated on both Word Error Rate (WER) and inference speed (RTFx), allowing for analysis of the accuracy-latency tradeoff for deployed voice systems.