Normalise Audio

Before you begin, you must first do a check wether the prompts were recorded correctly. When a prompt is not recorded properly, you often see that there are peaks in the volume (clipping). This can be investigated as follows:

  1. Open the supplied WAV file in Audacity
  2. You can immediately see whether it is a stereo (two channels) or mono (1 channel) recording
  3. When the audio already has peaks above 0.7 (see example below) then you can almost be sure that they are not properly recorded. Contact your voice over agency to get it sorted out.


  1. Additionaly you can watch the clipping, go to View -> Show Clipping
  2. If no red lines appear all is well. When there are red lines visible, Compass will make errors while converting the audio file, resulting in bad quality prompts. An example of a bad recording is shown below.


To ensure that you have prompts that always have the same sound level, you have to normalize them. To ensure an even better recording you also have to remove background noise. The whole process works as follows:

Record and normalize audio

  1. Set up a voicemailbox to which anyone can call to and set your own email address to mail the voicemail recordings to.
  2. Sit in a quiet room
  3. Take a phone with which has an HD horn
  4. The voice over actor can now call (for each prompt) to the previously set up voicemail number.
  5. Make sure that at least three seconds of silence is placed before you start the prompt. So start talking three seconds after the beep
  6. Speak out your prompt text clearly, make sure you talk calmly and avoid different volumes and tone of voices.
  7. Write down each recording attempt with a number (1, 2, 3, etc.) and write donw on whether they are successful or failed.
  8. Download the prompt from your mail and open it in Audacity
  9. Go to Effect -> Normalize
  10. Select: Remove any DC offset
  11. Select: Normalize maximum amplitude to: 0.0 dB (You can also put another number to make the sound louder or softer, the important thing is that you always use the same number here)
  12. Click on OK

Remove noise

  1. You can see that the entire line is stretched. It is also clear that silence is not true silence so now we have to strip it away:
  2. Select approximately 2 seconds of noise (found in the three seconds of silence before the prompt)
  3. Go to Effect -> Noise Removal -> Get Noise Profile
  4. Now select the entire prompt
  5. Go to Effect -> Noise Removal. Make sure the settings in Step 2 are as follows: Noise reduction (dB): 24 Frequency Smoothing (Hz): 150 Attack/decay time (secs): 0.15
  6. Click on OK

Remove silence

  1. Click on OK and you will see that the noise is replaced with silence
  2. Finally, you can shorten or lenghten (by copying and pasting silence) the silences.
  3. Ensure furthermore, that sounds like inhaling before a word is spoken, will also be replaced by silence
  4. When you have done all that you can remove the silence before and after the prompt by selecting it and pressing the delete button
  5. Then go to File -> Export and save the file as a WAV (Microsoft) signed 16 bit PCM


Have more questions? Submit a request


Please sign in to leave a comment.