Clone a voice from a recording

A voice is the audio source for an Odrazio project. Create one from a recording you are authorized to use, then wait for cloning to finish before selecting its language sample in a project.

Only clone your own voice or a voice for which you have valid, documented permission. A publicly available recording is not permission by itself.

Choose a clone type

• Multilingual is intended to create usable samples across the platform's supported languages. Its reference sample can be up to 10 seconds long.

• Single voice is intended for one source language and prioritizes that language's quality. It requires a clear 15–25 second sample.

• The creation page displays the current credit cost for the selected clone type and marks a first eligible clone as free when that entitlement is available.

Record a sample in the browser

1. Open Voices and select Create voice. Choose Multilingual or Single voice, then select Record.

2. Choose the source language and, when available, the desired microphone. Allow browser microphone access if prompted.

3. Read the reference text shown on the screen naturally and at a steady pace. For a single-language clone, read the longer prompt and meet the 15-second minimum.

4. Stop the recording, listen to it, and record again if there is background noise, clipping, or an error.

5. Enter a name and select Create voice.

Upload an existing sample

• Supported upload formats are WAV, MP3, and M4A.

• For multilingual cloning, upload a sample no longer than 10 seconds. For single voice, use a 15–25 second sample.

• Enter the exact words spoken in the file in the reference-text field. Do not summarize or correct the words; accuracy helps the system interpret the sample.

Make the source audio useful

• Record in a quiet room without music, echo, or other speakers.

• Use a natural, steady pace and keep the microphone roughly 15 cm from the speaker's mouth.

• Avoid whispering, shouting, rushed delivery, long silences, heavy effects, and recordings taken from a speaker.

• Use the same person throughout the recording; do not mix multiple voices.

After creation

The voice card shows whether it is pending, cloning, ready, or failed. A ready multilingual voice can expose language-specific samples; open it in the project voice selector and choose the sample language that matches the script. If a clone fails, correct the input issue and create a new sample rather than treating a failed asset as selectable output.