Back to browse
GitHub Repository

Benchmark transcription APIs against real meeting audio. Measure WER, diarization, latency, and cost.

3 starsPython

Open-source benchmark for transcription APIs on meeting audio

by eyepaqio·Apr 3, 2026·2 points·0 comments

AI Analysis

●●SolidSolve My ProblemNiche Gem

Tests Deepgram and AssemblyAI on actual crosstalk instead of clean audiobook samples.

Strengths
  • Targets meeting-specific artifacts like crosstalk and screen-share audio bleed in real tests.
  • CLI skips missing API keys instead of crashing on configuration errors.
  • Measures cost per hour alongside technical metrics for real budgeting decisions.
Weaknesses
  • Only supports four adapters; missing Google Cloud Speech and Azure AI.
  • Requires manual ground-truth JSON setup for custom audio samples.
Target Audience

Backend developers, CTOs choosing transcription infrastructure

Similar To

Hugging Face Spaces · MLPerf

Similar Projects