What we learned capturing conversations on noisy retail floors
Background music, overlapping voices and regional accents. Building conversation intelligence for real stores taught us a lot.
By the OrySense team · · 7 min read
OrySense started in consumer-durables showrooms and mobile stores. Those environments are about as far from a quiet call centre as you can get.
The audio problem
Store conversations overlap. Music plays. Customers bring family members who join in. Many conversations switch between languages. Off-the-shelf transcription models struggled badly, sometimes inventing text in quiet moments.
We benchmarked several speech models specifically on noisy retail audio and regional accents, and paired the one that performed best with dedicated speaker diarization — so the rep and the customer are separated even when they talk over each other.
The workflow problem
Store staff need both hands. That’s why OrySense supports one-tap recording in the app and hands-free capture with the Ory wearable, with audio synced after the conversation.
What managers needed
- A score per pitch they could compare across stores
- Which accessories, warranties and financing offers were missed
- A short summary instead of hours of audio
