· 6 min read
How to audition a voice
Naturalness is the wrong thing to audition for. What to listen for instead, when the voice you pick will speak on thousands of calls.
by Woise Team
Choosing a voice is the part of building an agent that feels like taste and mostly is not. There are a few things that reliably matter and a lot that reliably does not, and the difference only shows up if you audition properly.
Audition on your script, never on the sample
Every voice sounds good on the demo sentence. The demo sentence was chosen because it sounds good. What you need to know is how the voice handles your material, which will contain things the sample did not: your product names, postcodes, times, prices, and whatever phrase your business says a hundred times a day.
Paste in the actual greeting and the actual confirmation line. A voice that is lovely on prose and strange on numbers is a bad fit for a booking agent, and you will not discover that from a sample about the weather.
Listen down a phone, not through good speakers
The call is going to arrive on a phone, through a narrow codec, possibly on speaker in a van. That path removes a lot of the detail that makes a voice sound impressive on a laptop.
Voices that depend on breathiness or fine texture tend to lose the most. Voices with clear consonants and a steadier register survive better. Audition through the medium you are shipping to and the ranking changes more than you would expect.
Pace matters more than timbre
Most complaints that get described as the voice sounding wrong turn out to be about speed. Slightly too fast reads as impatient and gets people asking for things to be repeated. Slightly too slow reads as condescending and gets people interrupting.
Pace also interacts with the content. A list of appointment slots needs more space around it than a greeting does, because the caller is being asked to hold three options in their head and choose.
Consistency beats character
A distinctive voice is enjoyable once. What you actually need is a voice that sounds the same on the four hundredth call as the first, and does not do anything surprising when the text contains an unusual name or an abbreviation.
Test the awkward inputs deliberately. Numbers read as digits against numbers read as words, times, dates, postcodes, and any word your industry pronounces differently from the general public. That last one catches more agents than anything else on this list.