Speech Event Detector: 5.8M Parameter Non-Verbal Sound Model
October 8, 2026
The Speech Event Detector is a convolutional-Transformer model that identifies 54 classes of non-verbal vocalizations and background sounds. It outputs probabilities every 40 ms, allowing for timestamped integration into Whisper-based transcriptions.
HOW THIS AFFECTS YOU
●
builderYou can enrich your speech-to-text products with metadata about laughter, breathing, and environmental noise.