How to clone a voice

Definition

Cloning a voice is a four-step flow on the Speechify Build API: record a sample, capture consent, create the voice, and synthesize with it. Zero-shot cloning makes this self-serve, so the whole path fits into an app or a script.

The steps

First, record a clean sample of 10-30 seconds, under 60 seconds and under 5MB, with no background noise. Second, create a challenge with POST /v1/voices/consent-challenges, sending the speaker’s full name in the JSON field full_name, and record the same speaker reading the returned phrase. Third, POST to /v1/voices as multipart/form-data with name, gender (male, female, or not_specified), the binary sample, consent_challenge_id, and binary consent_recording. The response’s id is the voice ID used for synthesis. Follow the full quickstart for the challenge lifecycle and retry behavior.

Synthesizing

Pass the created voice’s id as voice_id to POST /v1/audio/speech, or the streaming endpoint, with your text. Check the voice’s models array before choosing a model. On simba-3.5 the voice speaks all 48 of its languages; set language to the language of your text. The older Simba 3.2 is English only, and Simba 3.0 covers 6 languages across 7 locales.

Consent is required

The current create call requires consent_challenge_id and consent_recording. The speaker must read the issued phrase, and Speechify checks the recording against the phrase and voice sample. The previous name-and-email consent field is deprecated. See the consent page and the voice cloning API.