HeyGen Voice Clone

HeyGen Voice Clone: Instant AI Voice Cloning API

What is HeyGen Voice Clone?

HeyGen Voice Clone is the instant voice cloning path of HeyGen Voice, HeyGen's in-house speech model. You send one clean recording of a single speaker and the API returns a private voice_id, usually ready within seconds, with no training step and no voice-clone slot to manage. That voice_id plugs straight into HeyGen Voice Text to Speech, so the cloned voice can read any script you give it later.

The model is built for speed and low friction. A 30-second clear sample is enough, one to three minutes is ideal, and only the first three minutes of a recording are used. The clone captures the speaker's tone, cadence, accent, and dialect, and the resulting voice stays private to you and your team.

Key Features

• •Instant clone from a single MP3 or WAV recording, returned as a reusable voice_id.
• •Similarity control (medium, high, extra_high, max) to set how closely the clone tracks the sample.
• •Multilingual cloning across 40 base language codes, including en, es, hi, fr, de, ja, and zh.
• •Optional gender and age hints, auto-detected when you leave them unset.
• •Built-in background-noise removal, enabled by default, for imperfect source audio.
• •Private voice_id scoped to your workspace, ready for HeyGen Voice text-to-speech.

Best Use Cases

Creators use HeyGen Voice Clone to build a signature narrator once and reuse it across hundreds of videos, courses, and podcasts without re-recording. Localization teams clone a presenter and read one script across many languages while holding the speaker's tone. Product and support teams give an assistant a consistent branded voice, and marketers personalize outreach at scale. In testing, a 43-second English sample cloned cleanly at similarity max and returned an active voice_id in about seven seconds.

Prompt Tips and Output Quality

Quality comes from the recording, not the text. Use one speaker, a quiet room, and clear natural delivery with some emotional variation. Aim for 30 seconds to two minutes of speech. Set language for reliable non-English clones, raise similarity to max for the closest match, and keep background-noise removal on for noisy input. Remember the output is a voice_id, not audio; generate speech with HeyGen Voice Text to Speech.

How to Use HeyGen Voice Clone

HeyGen Voice Clone turns one clean recording into a reusable voice_id for HeyGen Voice Text to Speech. Send two required fields, name and audio_url, and read the voice_id from the JSON response. The whole clone usually completes in seconds, so you can create a voice on demand inside a larger pipeline.

Getting a good clone

The recording drives everything. Provide one speaker, recorded in a quiet room, speaking naturally with a little emotional variation. Host the file at a public MP3 or WAV URL. Thirty seconds works, one to three minutes is ideal, and only the first three minutes are read. Avoid long silences and background music, and keep remove_background_noise on when the source is less than perfect.

Recommended Settings

These values were tested end to end and returned an active voice_id in about seven seconds from a 43-second English sample.

• •name: a clear label such as Clay Studio Narrator (EN) or Host. 1 to 64 characters.
• •audio_url: a public MP3 or WAV of one clear speaker, 30s to 2min.
• •similarity: max for the closest match to the sample. Drop to high or extra_high if the clone sounds over-fit or unstable.
• •remove_background_noise: true. Leave it on unless your source is already studio-clean.
• •language: set it (for example en, es, hi) for reliable non-English clones; leave unset to auto-detect.
• •gender and age: optional hints (male/female, young/old). Leave unset to auto-detect, or set them to steer an ambiguous sample.

Settings by use case

• •Signature narrator: similarity max, background-noise removal on, one to two minutes of varied speech for a stable, expressive voice across many reads.
• •Multilingual localization: set language to the sample's language so pronunciation tracks correctly, then reuse the voice_id across scripts.
• •Noisy or phone-quality source: keep remove_background_noise on and prefer a longer sample; set gender and age if auto-detection guesses wrong.

The response is a voice_id, not audio. Pass it to HeyGen Voice Text to Speech to generate speech; an instant voice cannot be retrained, so create a new voice when you need to change it.

How much audio does an instant voice clone need?

Thirty seconds of clear speech is enough. One to three minutes is ideal, and only the first three minutes are used.

How long does cloning take?

An instant voice is usually active within seconds of the request.

What does the API return?

A private voice_id, scoped to you and your team, that you pass to HeyGen Voice Text to Speech.

Can I retrain an instant voice?

No. An instant voice cannot be retrained; create a new voice to change it.

Which languages are supported?

Cloning covers 40 base language codes such as en, es, hi, fr, de, ja, and zh, auto-detected when language is unset.