Configure TTS voices
Every Say Message node in your IVR — and the Customer-first greeting on each queue — uses text-to-speech to read the text you typed. The voice that speaks it is configured per language on the Voices tab.
Before you start
- A phone number provisioned with at least one IVR flow assigned.
- A decision on which languages your callers speak. Most channels need only one; multi-language operations need a voice per language.
Steps
- Open Settings → Voice.
- Select your phone number in the sidebar, then go to the Voices tab.
- Pick the language you want to configure from the language picker.
- Pick a voice from the catalog for that language. Each voice has a sample button — preview it before saving.
- If the preview shows an amber warning or notice, the requested voice was coerced to a Twilio-renderable fallback. Preview uses the same compatibility logic as live calls, so the saved voice should match what callers hear.
- (Optional) Adjust speech rate and pitch if the catalog exposes those controls for the selected voice.
- Repeat for each additional language you serve.
- Save.
The IVR’s Say Message nodes and Customer-first greetings now speak in the chosen voice when the call’s language matches.
How the call’s language is decided
A call’s language is determined in this order:
- If the IVR has a Language Selection node and the caller picked an option, the call’s language is set to that choice.
- Otherwise, the call uses the channel’s default language as configured on the Voices tab.
You can read or set the call’s language mid-flow via a Set Variable step (using the call’s language variable) — useful when an API call earlier in the flow returned the caller’s preferred language.
Verify it worked
Dial your number and walk through an IVR step that includes a Say Message. The message should be spoken in the voice you configured for that language.
If the voice sounds wrong, double-check that the call’s language matches the language whose voice you edited — if you only configured an English voice but the call ran in French, you’ll hear a fallback.
If a preview or live call uses a fallback, Atender applies the same Twilio compatibility checks in both places. Watch for an amber warning or header-driven notice in preview that explains when the requested voice was coerced to a renderable voice.
Catalog notes
- The voice catalog covers 40+ languages.
- The catalog includes Twilio-renderable Amazon Polly voices plus Google Chirp3-HD voices where Twilio supports them.
- Each language typically has both male and female voice options.
- Some languages have neural or Chirp3-HD voices that sound more natural; others only have standard voices. Preview before committing.
- The voice is fixed at the channel level — you can’t have different Say Message nodes in the same flow speak in different voices unless you switch the call’s language.
Troubleshooting
-
Symptom: The voice sounds robotic and you remembered it sounding better in another tenant. Fix: The other tenant likely picked a neural voice. Check the Voice quality indicator next to each option in the dropdown and pick a Neural variant if available.
-
Symptom: A Google Chirp3-HD voice is not available for the language or locale you need. Fix: Choose one of the listed Amazon Polly voices for that language instead. Don’t assume every Chirp3-HD voice can be used in every locale — the catalog only shows voices that Twilio can render for that language.
-
Symptom: A Say Message reads numbers wrong (e.g. “twelve oh five” instead of “twelve-oh-five” for 12:05). Fix: TTS engines vary in how they interpret formatted numbers. Spell out tricky values in the Say Message text — “twelve oh five” reads more reliably than “12:05”.
-
Symptom: You can’t pick a voice for a language you need. Fix: That language isn’t in the catalog. Use the closest related language as a workaround (for example, Brazilian Portuguese for European Portuguese), or upload a pre-recorded Play Audio node for that specific message.