Voices

Pick any voice ID from the tables below and set it on session.output.voice in a session.update before session.ready. session.output is immutable once the session is established, so the voice can’t be changed mid-conversation.

1
{
2
  "type": "session.update",
3
  "session": {
4
    "output": { "voice": "ivy" }
5
  }
6
}

Language support

The voice agent’s input (speech recognition) and output (speech synthesis) cover different sets of languages:

• Input (understood): 🇺🇸 English, 🇫🇷 French, 🇩🇪 German, 🇮🇹 Italian, 🇵🇹 Portuguese, and 🇪🇸 Spanish.
• Output (spoken): those six, plus 🇮🇳 Hindi, 🇯🇵 Japanese, 🇰🇷 Korean, 🇨🇳 Mandarin, and 🇷🇺 Russian.

The agent can speak a language it can’t transcribe from user audio. This is useful for translation-style flows where the user speaks one of the recognized languages and the agent replies in another.

Choose a voice by language

Every voice supports every output language. The difference between the two tables is the voice’s primary accent:

• For an English accent (American or British) carried into other languages, pick from Voices.
• For a native accent in a specific non-English language, pick the matching language-specific voice.

Voices

These voices have an American or British English accent. They speak 🇺🇸 English, 🇫🇷 French, 🇩🇪 German, 🇮🇹 Italian, 🇵🇹 Portuguese, 🇪🇸 Spanish, 🇮🇳 Hindi, 🇨🇳 Mandarin, 🇷🇺 Russian, 🇰🇷 Korean, and 🇯🇵 Japanese. Their English accent carries over into the other languages.

Language-specific voices

These voices have a native accent in a specific non-English language. They also speak 🇺🇸 English, 🇫🇷 French, 🇩🇪 German, 🇮🇹 Italian, 🇵🇹 Portuguese, 🇪🇸 Spanish, 🇮🇳 Hindi, 🇨🇳 Mandarin, 🇷🇺 Russian, 🇰🇷 Korean, and 🇯🇵 Japanese, and they code-switch naturally between their primary language and English.