How to train a Replica Voice

Process

  1. Create a new voice via the 'Voices' page.
  2. Go through the 'Voiceprint' creation process.
  3. Download the relevant Replica Script for the voice you're creating (Voice-over, Character or Narration)
  4. Record the script
  5. Send us the audio files
  6. We train your Replica Voice

Requirements

Adhering to the following requirements will greatly improve the chances of producing a high-quality Replica Voice.

Source audio requirements

  • The only voice in the files uploaded should be the speaker being replicated. It's very important that any other speakers are trimmed out, otherwise, the generation will fail.
  • No background music however slight. Please ensure that the uploaded files have been cleaned up to removed any music or extra noise.
  • Please ensure you leave 1 second of Silence between any takes, or if you are joining separate tracks together.
  • For the sake of consistency, record all your audio using the same microphone, interface, and computer setup.
  • Minimal post-processing or EQ is preferred. However, we have seen decent results with highly processed audio
  • Record one file per script and save each file as the name of the script.
  • Export quality of 44.1KHz
  • File format of .wav
  • You can upload multiple files per speaker with a max file size of 500MB

Recording requirements

Use a studio-grade condenser microphone

It is imperative that you use a studio-grade condenser microphone. These microphones generate low self-noise and are capable of capturing wide dynamic range so they reproduce all the tonal qualities of your voice.

Use a pop filter

While not required, it is highly desirable to use a pop filter when recording. Pop filters capture puffs of air that are expelled from saying certain consonants. For example 'Pop' has a strong 'P' sound at the start of it and if not using a pop filter this air would go straight into the microphone and would be recorded as a thud-like sound.

Pop filters are relatively cheap, however, you can make yours from cheap materials found in your home. One method is to bend a wire coat hanger into a rough circle, wrap stocking or sock over it and tape it together so the stocking stays in place. Then, simply hold it between your mouth and the microphone as your record.

Distance from microphone

Your microphone should be at around 4 inches (10 centimeters) from your mouth. This will ensure that your voice is crisp, smooth and clear.

You do not have to speak too loud in order for the microphone to pick up your voice.

If you are too far from your microphone, it will be very noticeably hard to hear what you're saying.

When recording alternate styles/emotions like angry or upset, your voice may naturally increase in volume. If this is the case, you may want to stand another 2-4 inches away from the microphone so the recording doesn't distort.

Use sound dampeners

Your microphone should be about 12 inches (30 centimetres) from your sound dampeners.

Position your sound dampeners (or make-shift sound dampeners) around your microphone so that when you speak your voice does not bounce off the walls and instead is absorbed by the sound dampeners.

Practice each passage before recording

Before you hit record, make sure you practice each sentence. Especially if you're recording a character voice which you intend to use in film, animation or games.

Practising each sentence allows you to tap into the emotion as you speak, otherwise, it may sound unnatural like you're reading straight from the script.

Speak in one style

It is important to speak in one style for each recording session. For example, if you're performing a voice that does not sound like your natural speaking voice, make sure you stay in character.

Additionally, do not swap between happy, sad or angry emotions while recording.

Doing either of the aforementioned will result in your Replica Voice delivering varied styles when it's being used.

The ability to record many different styles (eg, happy, sad, angry) for your Replica Voice is coming soon.

Speak clearly

Be sure to enunciate each word as best as you can. Understandably, depending on your natural accent or the accent of the character you're performing, some words may not sound how they're spelt. But for the most part, make sure you speak clearly.

Do not improvise

If you are using one of our provided scripts, do not improvise. This will make it hard for the AI to transcribe what you're saying and match it with the script.

Improvising and not sticking to the script will result in less usable audio, thus lowering the quality of your Replica Voice.

Recording software

You don't need expensive software in order to record. Both Mac OS and Windows have free options. If you're not sure where to begin, here are a couple of suggestions.

Garageband for Mac

If not already installed, Garageband is available in the Mac App Store for free. It's relatively simple to use and there are many tutorials on YouTube.

Audacity for Windows

Available on Windows, Linux and Mac OS, Audacity is an open source, bare bones recording software and is the most used free recording software available. Just like Garageband, there are plenty of tutorials on YouTube for Audacity.

Got more questions? Message us in the app or email us at contact@replicastudios.com