FAQ for Voice Cloning

The AI needs lots of source material to make a good clone - 2 / 3 hours of recording is best

Use a script that matches the emotional style you want the voice clone to have - e.g. if the voice clone will be used for radio advertisements, and you want it to have an exaggerated commercial style, you could use a script that is a combination of lots of radio ads. You could get ChatGPT to write this, or ask Audiostack for help if you need it.

Read the script in the style that you want the voice clone to sound like. E.g. am exaggerated commercial voice, e.g. a calm steady voice, etc.

Building on the point above - if you want an commercial voice for advertisements, be extremely exaggerated in the recording. The AI will tone it down. Otherwise there is a risk that the clone sounds more “flat” than you want it to be.

Record in a quiet environment to minimise background noise. Avoid recording in an echoey room with lots of hard surfaces.

Enunciate words clearly and maintain a consistent volume and distance from the microphone.

Position the microphone close to the speaker's mouth (but not so close that the audio clips)

Use a pop filter. This is a small mesh screen you can place in front of the microphone to reduce the risk of certain sounds (such as "p" and "b") causing clipping.

If your microphone or interface has controllable input levels, aim to set your levels so that you're staying around the top end of the "green" range.

Use a decent microphone (generally condenser microphones pick up greater detail for vocal recordings) for better sound quality, or consider renting a studio environment for an hour if you're on a budget.

Remove breathing sounds as much as possible from the audio file before sending it to Audiostack

Do not apply any audio effects e.g. compression to the file, leave it completely RAW

Send the file to Audiostack in .wav format, with the largest file size possible to ensure good audio quality