Every call passes through numerous computers between the speaking parties. Why do you think the largest best-funded organizations in the world would lack the computational power to "listen"?
Usually telecom equipment is near the lower bound of the computer power needed to carry calls; there is not much room for advanced signal processing, speech-to-text, etc.
They are not getting through a single centralized location, and transcoding voice traffic does not require as much performance as voice recognition (the difference is several orders of magnitude). Your average telecom branch exchange station is not anywhere in the ballpark.