My voice model broke 20 minutes before a podcast in Austin, so I read the script myself
I had a cloned voice set up to read sponsor spots for a show I record in my garage, and it started spitting out garbled syllables 20 minutes before we hit record last Tuesday. I could not fix it in time, so I just read the whole thing live. Are we too quick to patch over these tools when they fail, or should we always keep a boring backup plan?
Man, I heard a guy on a tech podcast say the same thing happened to him and he just rolled with it. His point was that live reads actually land better with listeners anyway, so maybe the backup plan should just be you all along. Seems like the tools are nice until they aren't, and a human voice never crashes 20 minutes before showtime.