Skip to main content
A voice agent does two things during a call: it listens to the caller, then it speaks. Both rely on a voice technology:
  • Speech recognition turns what the caller says into text that the agent can understand. It is also called transcription.
  • Text-to-speech turns the text written by the agent into a voice. It gives the agent its tone and its way of speaking.
This page covers every voice setting, found in the Voice and languages tab of your agent. They are listed in the order they appear on screen.
Voice and languages

Spoken languages

Define the languages your agent can speak.
1

Add languages

Click Add a language and choose each language your agent needs. Each language includes its regional variant, for example “French (France)”.
2

Identify the main language

The first language you add carries the Main badge. It is the language used by default when the agent cannot determine the caller’s language.
An agent can speak up to 5 languages. The counter next to the button (for example “1/5”) shows how many languages are already configured.

How the agent picks a language

The agent detects the caller’s language automatically. The caller does not need to say which language they speak.

Detected language

If the caller speaks a language from your list, the agent answers in that language.

Unrecognized language

If the detected language is not in your list, the agent falls back to the main language.
If the agent picks the wrong language:
  • Check that the caller’s language is in the list.
  • Keep in mind that reliable detection needs a few words: a very short utterance (“yes”, “hello”) is not always enough.

Voice

Choose the voice your agent uses to speak. It has a strong effect on your brand image: a calm, neutral voice does not come across like a warm, energetic one. The list shows every voice available for your languages. To find the right one faster:
  • Use the search bar to find a voice by name.
  • Filter the Data, Gender (man, woman or neutral) and Provider columns.
  • Select a voice with the radio button to the left of its name.

Voice providers

Voices come from several text-to-speech providers: Google, ElevenLabs, Azure, Cartesia and Gradium. Each has its own catalog and its own sound. Listen to several voices before choosing, because two voices from the same provider can sound very different.
Voices with “Multilingual” in their name are designed to speak several languages. They are useful if your agent must answer in French and English, for example, with a consistent voice.
The Data column shows where the voice data is processed (for example “Europe”). It helps if your compliance requirements call for European hosting.

Speech rate

The slider sets how fast the agent speaks, from Slower to Faster, with Normal in the middle.
Speech rate slider

Slower

For complex information (numbers, addresses) or for callers who need more time to follow.

Normal

The standard pace, calibrated for natural conversation.

Faster

To deliver information quickly to callers in a hurry or used to the service.
The right setting depends on your audience and on what the agent has to say. Try several positions with the voice preview below before deciding.

Voice preview

Voice and languages
The voice preview lets you listen to the voice with the current configuration, without placing a call.
  1. Type a sentence in the “Type text to listen with current voice settings” field.
  2. Click the speaker icon to play it.
The preview uses the voice and speech rate you just chose: change them, then replay the same sentence to compare. Use a sentence the agent will actually say, with a brand name or a number, to spot pronunciation issues.

Background audio

Background audio adds a discreet background sound during the call to make the conversation feel more natural. A call that is perfectly silent between sentences can feel artificial. Choose the sound from the dropdown. The default is None.

Advanced settings: pronunciation

Synthetic voices pronounce common words very well, but they struggle with brand names, acronyms and technical terms. Three settings fix these cases, under Advanced settings - pronunciation. Each has its own tab.

Phonemes

Voice and languages
A phoneme is a sound. This tab lets you describe the exact pronunciation of a word by writing its sounds, so the voice says it the way you want. It is useful for:
  • proper names and brand names;
  • technical terms;
  • acronyms and abbreviations;
  • foreign words.
Phoneme configuration is only available for agents that use an Azure or Cartesia voice.
1

Add a phoneme

Click Add a phoneme.
2

Define the word and its pronunciation

Fill in both fields:
  • Word: the word to correct, as it is written.
  • Phoneme: its pronunciation, written in the International Phonetic Alphabet (IPA).
A phoneme is specific to a language. The same word can be pronounced differently in French and in English: configure it for each language of your agent.
After each addition, place a test call or use the voice preview to check the result. Start with the words the agent says most often.

Homophones

Voice and languages
Homophones are words that sound the same but are spelled differently, such as “made” and “maid”, or “peace” and “piece”. This setting is about listening, not speaking: the agent hears a sound, and speech recognition has to guess the matching word without any visual clue. When two words sound alike, the transcription can pick the wrong one, and the agent misunderstands. Configuring a homophone tells the agent: “in my business, when you hear this sound, it’s this word.” It helps you to:
  • reduce transcription errors;
  • improve the accuracy of the agent’s answers;
  • tailor recognition to your industry’s vocabulary.
1

Add a homophone

Click Add a homophone.
2

Define the word to prefer

In the Homophone field, enter the word the agent should keep in your context.

Best practices

  • Spot recurring errors in your call transcriptions before adding rules.
  • Focus on words specific to your industry.
  • Update the list as your vocabulary evolves.

Voice substitutions

Voice and languages
A substitution replaces some text with other text just before the voice speaks it. The displayed text does not change: only the way it is said. It is the simplest setting for abbreviations, symbols and acronyms that the voice reads poorly.
  • Units: “cm” becomes “centimeter”, “kg” becomes “kilogram”.
  • Symbols: ”+” becomes “plus”, ”=” becomes “equals”, ”>” becomes “greater than”.
  • Titles: “Dr” becomes “Doctor”, “Mr.” becomes “Mister”.
  • Addresses: “St” becomes “Saint”, “Av.” becomes “Avenue”.
  • Acronyms spelled out: “SNCF” becomes “S N C F”.
  • Acronyms expanded: “VAT” becomes “Value Added Tax”.
  • Technical terms specific to your industry.
  • ”€” becomes “euros”, ”@” becomes “at”, ”&” becomes “and”.
  • Regular expressions let you create advanced rules for complex cases.

Add a substitution

1

Open the form

In the Substitutions tab, click Add a substitution.
2

Fill out the form

Use the per-language option when an abbreviation is said differently depending on the language. For example, “cm” becomes “centimeter” in English and “centimètre” in French.

Best practices

  • For an ambiguous acronym, prefer spelling it out (“IBM” becomes “I B M”).
  • Keep only the substitutions you really need: the more there are, the harder the setup is to maintain.