Fact · Speech recognition
8hours of audio
Artificial Analysis tested its streaming speech-to-text models on about 8 hours of audio, the same audio as its offline benchmark.
Checked against its source on October 17, 2026
In the source's own words
“we test these streaming models on the same ~8 hours of audio as our offline benchmark”
Announcing AA-WER Streaming (Artificial Analysis), Artificial Analysis, Artificial Analysis, 2026, Opening paragraph of the article.
Open the original- What it measures
- The benchmark measures latency and word error rate after the Silero voice activity detector reports end of speech, and reports first partial and first final results separately.
- What it does not show
- Eight hours is a useful sample but still one benchmark on one set of recordings, and it does not describe a particular room or microphone.