What is text-to-speech?
- Text-to-speech
- Text-to-speech converts written words into spoken audio, and on an AI phone call it is the final step: the voice the caller actually hears.
What text-to-speech means in practice
Modern synthesis has moved a long way from the flat, clipped delivery of old phone systems. Pauses and emphasis now land roughly where a person would put them.
Where it still slips is the unusual word. A surname, a street name or a local pronunciation can come out wrong in a way that is immediately noticeable.
Speed matters as much as quality. A voice that is technically beautiful but arrives late feels less natural than a plainer one that answers promptly.
What people get wrong
How a voice reads "$149" and "Dr."
Before any sound is made, a speech engine has to turn your text into words, and that step is where odd things happen. Take the line, "Dr. Patel's office is at 1200 Oak Dr., St. 4. Cleanings are $149." A person reads "Dr." two different ways without thinking. An engine has to guess each time, and "St. 4" might come out as "Saint four" or "Street four" when you meant "Suite four".
Numbers carry the same risk. "1200" can be "twelve hundred" or "one thousand two hundred", and a phone number typed as 5550142 may be read as five million and change. Times such as "8-5" can come out as "eight minus five". None of this is a voice quality problem. The most natural voice you can buy will still say "Saint four" if the text lets it.
Proofreading with your ears
Every line the assistant can speak deserves one listen, and that's less work than it sounds. Write numbers the way you'd say them, so "one forty-nine" if that's how your front desk says it. Phone numbers go in groups with commas, such as "555, 01, 42". Spell out "Suite", "Drive" and "Doctor" so there's nothing to guess.
Listen on a phone, not on laptop speakers. Phone audio drops the high and low ends of the sound, so a word that's crisp on your computer can smear on a call, and names with soft consonants suffer most. When a name comes out wrong, respell it the way it sounds ("Nwin" for Nguyen, if that's how your colleague says it) and listen again. Keep a short list of those respellings somewhere your team can find it, because a new hire's name will need adding one day.
How GreetKeeper handles it
You write every line GreetKeeper speaks, which is where most of the improvement is available. Plain sentences that name your business read better than anything formal.
The greeting can disclose that the caller is speaking to an assistant, which Maine's chatbot law effectively requires and Utah requires verbally in regulated occupations.
We make no claim that the voice is indistinguishable from a person. Hear it on a demo with your own scenario and judge it yourself.
Voice questions
Can I use a clone of my own voice?
Cloning an identifiable person's voice raises consent and impersonation questions well beyond a product setting, and it is not something to do casually on a business line.
How do I fix a mispronounced name?
Spell it phonetically in the text. Writing a name the way it sounds rather than the way it is spelled fixes most cases in a single edit.
Does a faster voice sound more natural?
Up to a point. Past it, callers start talking over the assistant, which makes the transcript messy and the call worse. Test rather than assuming faster is better.
Related terms
Hear it take one of your calls
Two minutes, your own scenario, no card.