Clone a voice and use it everywhere

A cloned voice is not a per-feature setting. Clone once and the same voice_id works in text to speech, in a realtime session, in live interpretation, and on a phone call.

Prerequisites

A paid plan and 100 credits

Voice cloning is available on Plus and above, and each new clone costs 100 credits — see Pricing.

A clean sample

A few seconds of one person talking, no music or second speaker underneath. The clone inherits whatever is in the clip, background noise included.

Recipe

import httpx

with open("sample.wav", "rb") as audio:
    created = httpx.post(
        "https://api.eesi.ai/v1/voices",
        headers={"Authorization": f"Bearer {EESI_API_KEY}"},
        files={"file": ("sample.wav", audio, "audio/wav")},
        data={"name": "Priya"},
        timeout=120,
    )
created.raise_for_status()
voice_id = created.json()["voice_id"]

Cloning is not instant. Poll until the voice is ready before using it:

import time

while True:
    voice = httpx.get(
        f"https://api.eesi.ai/v1/voices/{voice_id}",
        headers={"Authorization": f"Bearer {EESI_API_KEY}"},
    ).json()
    if voice["status"] in ("ready", "failed"):
        break
    time.sleep(2)

if voice["status"] == "failed":
    raise RuntimeError(voice["error"])

Hear it before you commit

POST /v1/voices/{id}/preview speaks a line you choose, so you can check a clone against real copy rather than the sample it was made from.

preview = httpx.post(
    f"https://api.eesi.ai/v1/voices/{voice_id}/preview",
    headers={"Authorization": f"Bearer {EESI_API_KEY}"},
    json={"input": "Your appointment is confirmed for Thursday at two."},
)
open("preview.mp3", "wb").write(preview.content)

Use it

The same id, three places:

httpx.post(
    "https://api.eesi.ai/v1/audio/speech",
    headers={"Authorization": f"Bearer {EESI_API_KEY}"},
    json={"model": "nur-tts-v1", "voice": voice_id, "input": text},
)

{
  "type": "session.update",
  "session": {
    "audio": { "output": { "voice": "ev_abc123" } }
  }
}

The session brief's `voice`, or the Care companion voice on the patient or the
organization.

In a realtime session the voice must go in session.audio.output.voice. The serving backend reads nowhere else, and a session without one drifts between voices mid-conversation.

Why it stays the same voice

Every voice — cloned or built-in — resolves to a stored reference clip, and that clip is sent with every request. Without one the model samples a new voice each time it generates, and a long piece or a long call wanders between speakers. This is why a clone sounds identical in part twelve and part one.

Build

Cookbook

Billing

Clone a voice and use it everywhere

Prerequisites

Recipe

Hear it before you commit

Use it

Why it stays the same voice

Going further

Voices

Realtime

​Prerequisites

​Recipe

​Hear it before you commit

​Use it

​Why it stays the same voice

​Going further

Voices

Realtime

Prerequisites

Recipe

Hear it before you commit

Use it

Why it stays the same voice

Going further