Benchmark LLMs on your own transcripts with Voice Standard Bench

General LLM leaderboards score coding, math and chat. None of them score what a model does with a real transcript: the filler words, missing punctuation and misheard names that come out of Speech-to-Text. Voice Standard Bench does. In this video we clone it, point it at a folder of our own audio, and get back a leaderboard for those transcripts. Voice Standard Bench scores every model on AssemblyAI's LLM Gateway, 50+ today, on the tasks voice products hand to an LLM: summarization, information extraction, text correction and dictation cleanup. Each task runs at four transcript lengths, from a