Semantic accuracy could be the future of ASR benchmarks

Most speech-to-text benchmarks don’t reflect reality. This one does. https://github.com/pipecat-ai/stt-benchmark?tab=readme-ov-file#results-summary Using real-world audio conditions and semantic word error rate, this benchmark shows what actually matters: understanding meaning, not just words. That’s the difference between: “renew my prescription” vs “review my prescription” One works. One doesn’t. Explore Speechmatics: https://www.speechmatics.com/speech-to-text Join the Discord community: https://discord.com/invite/speechmatics Try it yourself (free): https://portal.speechmatics.com/signup Voice AI agents: https://www.speechmatics.com/use-cases/ai-voice-agents GitHub (get started fast): https://github.com/speechmatics/speechmatics-academy Follow us: https://www.linkedin.com/company/speechmatics/ https://twitter.com/speechmatics https://www.tiktok.com/@speechmatics