The biggest problem I see with this is the scenario where calls are recorded. They’re recorded in case we hit a “he said, she said” scenario. If some issue were to be escalated as far as a courtroom, the value of the recording to the business is greatly diminished.
Even if the words the call agent gets are 100% verbatim, a lawyer can easily argue that a significant percentage of the message is in tone of voice. If that’s lost and the agent misses a nuance of the customer’s intent, they’ll have a solid case against the business.
I see no problem: they can record the original call and postprocess it with AI live for the operators. The recordings would be the original audio.