A Frankenstein's voice

I've been working my way through a number of telephone voice recognition systems this morning, paying some bills. I can see these systems growing uncontrollably, getting patched by different voice artists, inserting parts of phrases as the system expands. It is eerie listening to the Sydney train system, where announcements are made up of sentence fragments stitched together by computer software.

I can see such a system getting to the size where it can no longer be re-recorded using a single speaker's voice as a consequence of our failure to handle natural language generation in software. Either because it is too hard to justify the expense, or the amount of recorded dialog exceeds the ability of any one person to record in a reasonable time. The consequence leaves companies to speak with a Frankenstein's collage of different voices, no one voice having authority over all others. There are parallels between this and the way businesses are becoming fractured, vitalized and spread globally.