Voices
Kalpa, Navana, Maya, Gemini and Soniox voices — their models, voices, the languages they speak, and every setting with its range and default.
These voices are chosen in Voice Settings → Voice Provider, in each language's editor in Languages, and in the flow builder's Voice & languages. Choose the vendor, then the voice; the vendor's own settings appear under it. A setting left empty uses the vendor's default.
They are offered where your platform can run them (see Voices and recognition). They can't be played from the dashboard yet: place a test call to hear one.
Kalpa
Conversational English and Hindi voices, including Hinglish.
| Model | Speaks |
|---|---|
| Kalpa TTS 1 (default) | English and Hindi, including Hinglish |
| Kalpa TTS beta | English only |
Voices. Kiara, Sana, Arjun, Aman, Nikhil and Dev speak Hindi and English. Wren, Tessa, Isla, Grant and Otis speak English. A Hindi agent is offered only the voices that speak Hindi.
| Setting | What it does | Range | Default |
|---|---|---|---|
| Temperature | Higher sounds livelier and more varied from line to line; lower sounds steadier. | 0 to 1.5 | 0.9 |
| Acoustic temperature | How much the voice's tone and texture vary, separately from what it says. | 0 to 1.5 | Kalpa's own |
| Audio quality | The quality Kalpa renders the voice at. | Low, Medium, High | High |
| Chunk schedule (More settings) | How much text Kalpa waits for before it speaks each part of a reply, in characters. A smaller first number starts the voice sooner. | 1 to 10 whole numbers, each 50 to 2,000 | 50, 80, 120, 150 |
Navana
Voices for nine Indian languages and English: Bengali, English, Gujarati, Hindi, Kannada, Malayalam, Marathi, Odia, Tamil and Telugu. Every voice speaks every one of them, as the same speaker.
| Model | Notes |
|---|---|
| Navana TTS (default) | Best quality |
| Navana TTS Fast | Faster, slightly lower quality |
Voices. Default female, Default male, and Navana's 55 named voices. To use a
voice that isn't listed, or your own cloned voice, enter its ID in Another
Navana voice. A cloned voice (an ID starting cv_) speaks only the language
it was cloned in.
Navana is told the language the call is in. When you change the agent's language, the voice follows it; if Navana doesn't speak the new language, the change says so and nothing is saved.
Five calls at a time
Navana speaks on at most 5 calls at the same time on your platform. For a campaign that places many calls at once, choose another voice.
Maya (preview)
Two voices for ten Indian languages and Indian English: Hindi, Bengali, Gujarati, Kannada, Malayalam, Marathi, Odia, Punjabi, Tamil, Telugu and English. Maya is marked Preview: try it on a test call before you use it on live calls.
| Model | Notes |
|---|---|
| Maya 2 Native | Indian languages and Indian English |
Voices. Ananya and Arjun. Both speak every language.
| Setting | What it does | Choices | Default |
|---|---|---|---|
| Language | Leave it to detect the language of each line (best for mixed languages), or fix one. | Detect from each line, or one of Maya's languages | Detect from each line |
Gemini
Thirty voices that pick up the language from the text, with a speaking style you describe in your own words. They speak English, Hindi, Bengali, Marathi, Tamil, Telugu, Arabic, Dutch, French, German, Indonesian, Italian, Japanese, Korean, Portuguese and Spanish.
| Model | Notes |
|---|---|
| Gemini 2.5 Flash (default) | The default |
| Gemini 2.5 Pro | Higher quality, slower |
| Gemini 3.8 Flash | Newer generation |
| Gemini 3.8 Flash Lite | Newer generation, lighter |
Voices. Puck (the default), Kore, Zephyr, Charon, Fenrir, Leda, Orus, Aoede, Callirrhoe, Autonoe, Enceladus, Iapetus, Umbriel, Algieba, Despina, Erinome, Algenib, Rasalgethi, Laomedeia, Achernar, Alnilam, Schedar, Gacrux, Pulcherrima, Achird, Zubenelgenubi, Vindemiatrix, Sadachbia, Sadaltager and Sulafat. The picker shows how each one sounds.
| Setting | What it does | Range | Default |
|---|---|---|---|
| Speaking style | How the voice should sound, in your own words, for example Speak warmly, a little slower than usual. | Text, at most 500 characters | Its natural style |
Soniox
Voices for sixty languages; every voice speaks every one of them. Among this dashboard's languages, Soniox speaks all but Assamese and Odia.
| Model | Notes |
|---|---|
| Soniox TTS v2 | Realtime |
Voices. Adrian (the default) and Daniel are listed. To use any other Soniox voice, or your own cloned voice, enter its name or ID in Another Soniox voice. Soniox is told the language the call is in, and follows the agent's language like Navana.
| Setting | What it does | Range | Default |
|---|---|---|---|
| Speed | How fast the voice speaks. 1 is its natural pace. | 0.7 to 1.3 | 1 |
| Reduce silence | Shortens the pauses within and between sentences. | On or off | Off |
Three calls at a time
Soniox speaks on at most 3 calls at the same time on your platform, and one reply can be at most 2 minutes of speech.
Speed and pronunciation
The Speed slider in Voice Settings is for the other voices: these vendors take their own speed (Soniox's, above) or none. Pronunciation rules apply to ElevenLabs and Cartesia voices only.