The voice agent sounds wrong or speaks the wrong language
Accent, language switching, mispronounced names and awkward pauses each have a different fix.
"It sounds wrong" covers several distinct problems, and the fix depends entirely on which one you have. Before diagnosing anything, listen to a recording — a second-hand report of what an agent sounded like is not diagnosable, and thirty seconds of audio usually is.
Before you start
- A recording to listen to. Start there rather than with the settings.
- Your plan tier to hand, since voice quality depends on it.
Steps
Listen to a recording first
On the contact's activity feed. You will hear things no description conveys: the agent talking over the caller, a pause that felt like a dropped line, a name said wrongly.
Check the voice is set for that language
Voice is set per language on the bot. A language with no voice configured will not sound the way you expect.
Try a different voice for that language
Every plan gets the same ultra-realistic voices, so a robotic result is not your plan and upgrading will not change it. It is the particular voice. Audition two or three saying your own opening line — the differences between them are much larger than people expect.
Check the language is enabled at all
A bot will not switch into a language it has not been allowed to speak. If callers are switching and the agent is not following, this is why.
Fix pronunciation in the instruction
Spelling a brand or place name phonetically in the instruction usually fixes it. This is a practical workaround and it works well enough to be worth doing rather than accepting a mispronounced company name.
Test the things you say every day
Prices, phone numbers, street names, appointment times. These are where pronunciation fails in ways that cost you bookings, and they take two minutes to check.
For long pauses, look at the model
A more capable model that answers more slowly can be a net loss on voice, because a pause on a call reads as the line going dead. Try a smaller, faster one.
Check it worked
Play a recording to somebody who speaks the language natively and ask one question: would you keep listening? That answer is worth more than any setting.
The parts people get wrong
- Nothing downstream of the bot can change how it sounds. Two markets needing two voices in the same language means two bots.
- Listening to three AI calls a week is the highest-return half hour available for improving a voice agent.
The ones people actually ask
What should I do before changing anything?
Listen to a recording. You will hear things no description conveys — the agent talking over the caller, a pause that felt like a dropped line, a name said wrongly that made the caller ask what.
It sounds robotic and everything is configured correctly.
It is not the plan — every plan has the same ultra-realistic voices, so there is nothing to upgrade to. Try a different voice for that language, listening to your real opening line rather than a sample. If what bothers you is timing rather than timbre, look at the model instead: a slower one produces pauses that read as a dropped line.
There are long pauses on calls.
Look at the model. A more capable model that answers more slowly is a net loss on voice, because a pause reads as the line going dead. Try a smaller, faster one.