Beyond the Tried voices, a workspace can have its own. Clone one from recordings of a speaker and it appears in the voice picker and in List voices, and any agent in the workspace can use it on the Premium or Standard tier.
Clone a voice#
Instant cloning creates a voice from a few short, clean recordings.
Open the dialog#
Click the voice button in the toolbar, then Clone a voice in the footer.
Name it#
Enter a Voice name, for example "Ava (front desk)".
Upload samples#
Under Audio samples, choose one to five recordings of a single speaker, MP3, WAV or M4A, at least a minute in total and at most 25 MB together.
Clone#
Click Clone voice. The new voice is selected in the picker. Click Use voice to put it on the agent.
Only clone voices you have permission to use.
Recording tips#
- One speaker, no music, no other voices.
- A quiet room and a close microphone. Background noise is cloned too.
- Speak the way the agent should sound on the phone: the same pace and energy.
Needs keys Cloning needs the platform's voice key. Without it, the voice is saved with a placeholder ID so you can try the flow, but it will not speak in that voice on real calls.
Clone with the API#
POST /clone-voice takes multipart/form-data:
| Field | Type | Required | Description |
|---|---|---|---|
voice_name |
string, 1 to 100 chars | Yes | Display name. |
files |
file, repeatable | Yes | Audio samples, 25 MB in total. |
description |
string, up to 500 chars | No | Notes about the voice. |
provider |
elevenlabs, fishaudio, minimax |
No | Default elevenlabs, which supports both voice tiers. Leave it out unless you need another engine. |
curl https://api.tried.com.au/clone-voice \
-H "Authorization: Bearer $TRIED_API_KEY" \
-F voice_name="Ava (front desk)" \
-F files=@sample-1.mp3 \
-F files=@sample-2.mp3The endpoint also accepts JSON for scripts: { "voice_name": "...", "files": [{ "filename": "a.mp3", "mimetype": "audio/mpeg", "content_base64": "..." }] }.
The response is a voice object. Cloned voice IDs start with custom_voice_. Set that ID as the agent's voice_id. See Clone voice.
Community voices (API only)#
For Retell compatibility, the API can also search a shared voice library and add a voice to your workspace. The dashboard does not show this library.
| Endpoint | Description |
|---|---|
GET /search-community-voice |
Query parameters search, gender, language, accent, page_size (1 to 100, default 30). Returns { "voices": [...] } with public_owner_id, voice_id, voice_name, gender, accent, language, description, preview_audio_url, use_case. |
POST /add-community-voice |
Body { "public_owner_id", "voice_id", "voice_name" } from a search result. Returns the workspace voice with a custom_voice_ ID. |
Added voices appear in the voice picker next to your cloned voices.
Limits and gotchas#
- Fallback voices in Security & Fallback Settings lists only the Tried voices. Cloned voices can be the primary voice but not a fallback. See Voice fallback.
- Previews need the platform's voice key; without one the browser reads a sample line in its own voice. Judge a cloned voice with Test Audio instead.
- A cloned voice's quality depends on the samples. If it sounds unstable, lower Voice Temperature under Speech Settings.