Show HN: Voice Standard Bench – Open benchmark of LLMs on speech transcripts

Category: ai-ml

Tags: benchmarking, llm, speech-transcription, go, evaluation

Score: 6.0/10 (Innovation: 5, Technical: 7, Documentation: 6, Utility: 6)

A Go CLI that benchmarks LLMs on speech transcription tasks (summarization, extraction, correction, dictation) using real AssemblyAI call transcripts, measuring latency and quality by audio-length buckets. It includes a separate mechanical scoring path for dictation and a streaming pass for time-to-first-token, with a canonical result set for incremental merging.

Target audience: AI engineers, ML researchers, speech API developers

Repository: https://github.com/AssemblyAI/voice-standard-bench · Go · MIT · 7 stars

View on Hacker News