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.