How we taught the system to hear a quiet customer at a gas station: bimodal analytics for service control

▤ Summary

The article explains how a two-modal analytics approach can help detect and understand quiet customers at gas stations. It combines speech recognition, speaker identification, voice activity detection, and re-identification to measure service quality more accurately. The piece compares ASR models such as GigaAM v3 and Whisper and shows how model tuning affects word error rate and ROI. It also describes a practical workflow for extracting service insights from audio and video data and exporting results into formats like JSON, Excel, and CSV.

▦ Linked Companies