Voices
Browse the voice catalog, preview a voice, clone your own and set a fallback.
Every agent speaks with one voice, set by voice_id. There is no default voice: you choose one when you create the agent.
Find a voice
GET /v1/voices returns the catalog. Your own cloned voices are listed first, then the platform voices by name. Filter with query parameters:
| Parameter | Matches |
|---|---|
provider | The speech provider behind the voice. |
gender, accent | Exact values. |
language | Voices whose language list includes this code, such as en or es. |
search | Part of the name, ignoring case. |
curl "https://api.dialgem.com/v1/voices?language=en&search=warm" \
-H "X-API-Key: $DIALGEM_API_KEY"The list is capped at 500 voices, so narrow it with a filter rather than paging. Put the voice's provider_voice_id in the agent's voice_id.
In the dashboard, the voice picker plays a short sample for each voice.
Clone a voice
Upload a recording of the voice and Dialgem creates a custom voice that only your organization can see. See Clone voice for the request, which takes an audio file as multipart form data.
Get consent first
Only clone a voice you have the right to use. Recordings of other people need their permission.
Delete a cloned voice with DELETE /v1/voices/{voice_id}. Agents still using it need another voice.
Speed and expression
| Field | Range | Effect |
|---|---|---|
voice_speed | 0.5 to 2 | Speaking pace. 1 is normal. |
voice_temperature | 0 to 1 | Higher is more varied delivery. Only ElevenLabs voices use it; it has no effect on Cartesia voices. |
pronunciation_dictionary | up to 200 entries | How to say names and terms. |
Fallback voice
Set fallback_voice_id to a second voice, to be used if the first cannot be used.
Test over a real call
The preview in the picker is a short sample. How a voice sounds on a phone line, and how soon it starts speaking, is best judged on a real call.
Pricing
The voice engine is one line of the per-minute price. See Billing and pricing.