bdharva / audio_cleanup

Shell script that uses SoX to clean up and normalize audio recordings

audio_cleanup

This script uses several Sound eXchange (SoX) effects in combination to normalise and trim voice recordings that may have been recorded using different microphones, with differing background noise etc.

Setup

• Install SoX (download or just use your favorite package manager)
• Locate the installation location for SoX and place the .sh script in that same directory
• Open terminal and add the following entry to your .bash_profile: alias cleanup="/usr/local/bin/cleanup.sh" (or replace "cleanup" with whatever you'd like to call it)

Usage

With the setup out of the way, you can now use the audio cleanup script with cleanup infile outfile, where infile is the audio file you'd like to clean up and outfile is how/where you'd like the output to audio file to be named and saved.

If you're working with a master file of multiple voice clips, you'll first need to break them apart. Record a list of the breakpoints you'd like to use and then repeatedly run sox infile outfile trim starttime =endtime on the source file to produce all of the output files you need for the individual audio clips.

Understanding the Script

You'll probably want to mess with some of the values in the script to tune it to your needs. Here's a brief overview of the SoX effects used. For more/better details, check out the SoX documentation.

• remix
  • Performs a mix-down of all input channels to mono
• highpass
  • Follows the form highpass n where n is a frequency in Hz
  • Applies a high-pass filter of the specified frequency
• norm
  • Alias for gain -n
  • Normalises the audio to 0dB FSD if no argument is provided
  • Arguments:
    • -n, where n is in decibels, will normalize to the specified amount below 0dB FSD
• compand
  • Follows the form compand attack1,decay1 soft-knee-dB:in-dB1,out-dB1,...,in-dBn,out-dBn gain initial-volume-dB delay
  • Attack and decay are in seconds, with attack typically less than decay
  • The soft knee dB value rounds the points where adjacent segments of transfer function meet
  • The list of input values must be in increasing order and are specified in dB relative to the maximum possible signal amplitude.
  • This is a great write-up that dives into the intricacies of compand in more clarity and detail than I will even bother to attempt here.
• vad
  • Voice activity detector
  • Attempts to trim silence and background sounds from the ends of the audio.
  • Can only trim from the front, so the audio must be reversed to trim from the back.
  • Options:
    • -T: time constant (seconds) used to help ignore short bursts of sound
    • -p: amount of audio (Seconds) to preserve before trigger point
    • -t: measurement level used to trigger activity detection
• fade
  • Follows the form fade n where n is a duration in seconds
  • Applies a fade effect of the specified duration to the beginning and the end of the audio clip
• reverse
  • Flips the audio clip (good for using vad to trim silence at the end and to flip back)