What it does
Cekura Bench compares voice agent platforms, realtime speech-to-speech models, production conversation quality, speech-to-text models and text-to-speech voices under documented test conditions. The benchmarks measure reliability, task completion, data accuracy, response time, voice quality, interruption handling and speech recognition accuracy, where each metric applies. Cekura Bench says it scores results on live calls, real call audio and agent sentences. Each benchmark explains its methodology. Each benchmark links to supporting reports, test cases or source material where available. Head-to-head comparison pages cover speech-to-speech and speech-to-text models tested under the same conditions. Pass³ is the share of scenarios where all three retained test calls passed. Cekura also runs simulated calls on your agent and scores every one.
Capabilities
- The speech-to-speech Agent performance benchmark covers 12 models and 3,198 calls.
- Turn Bench covers 13 speech-to-speech models across 6 line conditions.
- The speech-to-text benchmark covers 22 models and 1,000 clips.
- The text-to-speech benchmark covers 27 models and 104 agent sentences.
- The voice platforms section covers agent performance across 9 platforms and 82 scenarios. It also covers voice quality across 9 platforms and 270 calls.
When to use it
- You need to compare voice agent platforms.
- You need to compare two speech-to-speech models or two speech-to-text models directly.
- You want to test your own voice agent with simulated calls that are each scored.
