Voices

Every voice has an id, like ev_d116d96b, and that id works everywhere: text to speech, Nur and Live translation. Your library has the built-in voices, shared by everyone, and the voices your organization designed or cloned.

curl https://api.eesi.ai/v1/voices \
  -H "Authorization: Bearer $EESI_API_KEY"

{
  "data": [
    { "voice_id": "ev_d116d96b", "name": "Nur", "status": "ready", "is_builtin": true },
    { "voice_id": "ev_1a2b3c4d", "name": "Dana", "status": "ready", "is_builtin": false }
  ]
}

To hear a voice without spending a synthesis, GET /v1/voices/{voice_id}/sample returns a short-lived link to its reference clip.

Design a voice

Describe a voice in words, in any language, and hear three takes of it. Each take is a different speaker reading the same line. Keep the one you like and it becomes a voice of your own.

curl https://api.eesi.ai/v1/voices/design \
  -H "Authorization: Bearer $EESI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"description": "A calm older man with a deep voice"}'

{
  "instructions": "male, elderly, low pitch",
  "text": "Sit down, take your time. Tell me what is on your mind.",
  "language": null,
  "name": "Harold",
  "takes": [
    { "take_id": "eyJvIjo3…", "audio": "UklGRi…", "content_type": "audio/wav", "seconds": 4.8 },
    { "take_id": "eyJvIjo3…", "audio": "UklGRm…", "content_type": "audio/wav", "seconds": 4.6 },
    { "take_id": "eyJvIjo3…", "audio": "UklGRk…", "content_type": "audio/wav", "seconds": 5.1 }
  ]
}

Field	Required	Notes
`description`	Yes	Who the voice is: age, gender, pitch, an English accent, a whisper. Up to 300 characters.
`text`	No	What the takes read. Without it, a line in character is written for them.
`language`	No	The language of the line, such as `en` or `ar`. Without it, the language of the description.

Each take’s audio is base64 WAV. Nothing is stored yet. To keep a take, send it back, exactly as it came, within an hour:

curl https://api.eesi.ai/v1/voices/design/keep \
  -H "Authorization: Bearer $EESI_API_KEY" \
  -H "Content-Type: application/json" \
  -d "{\"take_id\": \"$TAKE_ID\", \"audio\": \"$TAKE_AUDIO\", \"name\": \"Harold\"}"

The kept voice works like a clone: it starts as processing, becomes ready within a few seconds, and speaks every language. No consent is needed because no one’s recording is in it. The takes are billed as speech, and keeping one counts as a clone. In the console, open Voices → Design from words.

Clone a voice

Upload a short, clean recording of one person speaking.

curl https://api.eesi.ai/v1/voices \
  -H "Authorization: Bearer $EESI_API_KEY" \
  -F file=@dana.wav \
  -F name=Dana \
  -F consent_attested=true

Field	Required	Notes
`file`	Yes	The recording. An audio file up to 10 MB and 60 seconds.
`name`	Yes	Up to 255 characters.
`consent_attested`	Yes	`true` confirms you have the speaker’s permission to clone their voice.
`language`	No	The language of the recording, such as `en` or `ar`.
`gender`, `accent`, `category`, `description`	No	Labels for browsing your library.

The voice appears right away with status: "processing", and becomes ready within a few seconds. Poll GET /v1/voices/{voice_id} until it does. Using it before then returns a 400 that says so.

Recording a good clip

The clone copies everything in the recording, not just the voice. Reverb, music and background chatter all come along. So:

3 to 10 seconds of connected speech is enough. Longer is not better, and over 60 seconds is refused.
One speaker, close to the microphone, in a quiet room.
Speak the way the voice will be used. A calm support voice from a calm reading, not from a shout.
Record in the language you’ll use most. A clone speaks every language, with a hint of the accent it was recorded in.

Hear it before you use it

POST /v1/voices/{voice_id}/preview speaks a fixed sentence and returns MP3, so every voice can be compared on the same line.

Rename or delete

curl -X PATCH "https://api.eesi.ai/v1/voices/$VOICE_ID" \
  -H "Authorization: Bearer $EESI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"name": "Dana (support)"}'

curl -X DELETE "https://api.eesi.ai/v1/voices/$VOICE_ID" \
  -H "Authorization: Bearer $EESI_API_KEY"

Only the name can change. Deleting erases the recording and everything made from it. Built-in voices can’t be renamed or deleted. A cloned voice that goes unused for three years is erased automatically. Every use resets that clock. A voice is personal. Only clone a voice with the speaker’s clear permission, keep a record of that permission, and tell listeners when they are hearing a synthetic voice. Every audio response from EESI is marked as machine-generated in its X-Synthetic-Audio header and file metadata.

Limits


Upload	10 MB and 60 seconds
Free plan	One clone. Add credits for more.
Errors	`400` for missing consent, a non-audio file, a voice still processing, or a take that expired or was changed; `404` for a voice your organization doesn’t own; `413` for a file over 10 MB

Get started

Quickstarts

Nur

Speech

Audio intelligence

Design a voice

Clone a voice

Recording a good clip

Hear it before you use it

Rename or delete

Limits

​Design a voice

​Clone a voice

​Recording a good clip

​Hear it before you use it

​Rename or delete

​Consent

​Limits

Design a voice

Clone a voice

Recording a good clip

Hear it before you use it

Rename or delete

Consent

Limits