“Eleven Labs' recently released Scribe v2 speech-to-text model achieves 93.5% accuracy on the Fleurs dataset across 30 languages with a latency of under 150 milliseconds, outperforming other models on benchmarks.”