Fact · Speech recognition
In a 2024 study of children's classroom dialogue, Google Speech-to-Text had the lowest mean error rate of the three engines compared and the lowest variation, which the authors describe as the most robust to noise.
Checked against its source on October 17, 2026
In the source's own words
“Google Speech-to-Text achieved the most robust performance in the face of substantial audio noise, with the lowest mean WER score of 0.78”
Automatically Detecting Confusion and Conflict During Collaborative Learning Using Linguistic, Prosodic, and Facial Cues, Yingbo Ma et al., arXiv, 2024, Table 2.
Open the original- What it measures
- The engines were Google Speech-to-Text, Rev.ai and OpenAI Whisper, and the audio came from elementary school learners working on programming tasks.
- What it does not show
- The paper reports its error scores on its own scale, which is not stated as a percentage. Children's classroom audio in one study is a narrow case, and the authors used Google transcripts for the rest of their work.