Sibme has upgraded our transcription pipeline for speech-to-text and speaker diarization (who spoke when).

What this means for you: More accurate transcripts, clearer speaker separation, and a stronger foundation for transcript-driven coaching and AI analytics—especially in real classroom audio where multiple voices and quick turn-taking can make attribution difficult.

Why it matters: Our new combination is designed to deliver higher-quality transcript and diarization signals to power Sibme’s AI Insights, AI Reports, and reporting.

AI Reports has been improved to interpret silence vs. group interaction reducing false “group talk” activity in Teacher/Student Talk Time reporting.

The parameters used to define group talk has been improved to more clearly define when there truly is multiple speakers vs. a conversation between teacher and students.

Question detection logic was improved to better match what was actually said in the video resulting in more trustworthy question analytics and cleaner distributions.

AI Reports logic for capturing and reflecting multiple hand-raise events was improved, including moments where several students raise their hands around the same time, reducing missed detections and improving confidence in participation analytics.