SoX (Sound eXchange)

Overview

Process audio with SoX — the Swiss Army knife of audio manipulation. Handles format conversion, effects (reverb, echo, EQ, compression, chorus), mixing/combining tracks, silence trimming, volume normalization, pitch/tempo changes, spectrograms, and batch processing. Lighter than ffmpeg for pure audio work, with a powerful effects chain syntax.

Instructions

Step 1: Installation & Basics

Install:

# Ubuntu/Debian
apt install -y sox libsox-fmt-all

# macOS
brew install sox

# Verify
sox --version

Basic syntax:

sox input.wav output.wav [effects...]
# or: sox [input-options] input [output-options] output [effects...]

Get audio info:

soxi file.wav
# Sample Rate: 44100, Channels: 2, Duration: 00:03:42.15

soxi -d file.wav    # Duration only
soxi -r file.wav    # Sample rate only
soxi -c file.wav    # Channels only
soxi -b file.wav    # Bit depth

Step 2: Format Conversion

# Format conversion (SoX infers from extension)
sox input.wav output.mp3             # WAV → MP3 (requires libsox-fmt-mp3)
sox input.wav output.flac            # WAV → FLAC (lossless)
sox input.mp3 output.wav             # MP3 → WAV (for editing)

# Sample rate, bit depth, channels
sox input.wav -r 16000 output.wav    # Downsample to 16kHz (for speech)
sox input.wav -b 16 output.wav       # Convert to 16-bit
sox input.wav output.wav channels 1  # Mix down to mono

# Raw PCM → WAV
sox -r 44100 -b 16 -c 2 -e signed-integer input.raw output.wav

Step 3: Trimming & Splitting

# Trim: keep from 0:30 to 2:00
sox input.wav output.wav trim 30 90    # start=30s, duration=90s

# Trim: keep first 60 seconds
sox input.wav output.wav trim 0 60

# Trim: skip first 5 seconds
sox input.wav output.wav trim 5

# Remove silence from beginning and end
sox input.wav output.wav silence 1 0.1 0.1% reverse silence 1 0.1 0.1% reverse

# Remove silence throughout (split into non-silent segments)
sox input.wav output.wav silence 1 0.5 0.1% 1 0.5 0.1% : newfile : restart

# Split into 30-second chunks
sox input.wav output.wav trim 0 30 : newfile : restart
# Creates output001.wav, output002.wav, ...

# Pad with silence
sox input.wav output.wav pad 2 3    # 2s before, 3s after

Step 4: Volume & Normalization

# Normalize to 0dB peak
sox input.wav output.wav norm

# Normalize to -3dB peak
sox input.wav output.wav norm -3

# Adjust volume
sox input.wav output.wav vol 1.5      # 150% volume
sox input.wav output.wav vol -6dB     # Reduce by 6dB
sox input.wav output.wav vol 3dB      # Increase by 3dB

# Dynamic range compression (make quiet parts louder)
sox input.wav output.wav compand 0.3,1 6:-70,-60,-20 -5 -90 0.2

# Limiter (prevent clipping)
sox input.wav output.wav compand 0.01,0.3 -80,-80,-6,-6,0,-3 0 0 0.01

# Loudness normalization (EBU R128-style with gain)
# First measure:
sox input.wav -n stat 2>&1 | grep "RMS lev dB"
# Then apply gain to reach target (-16 LUFS for podcasts):
sox input.wav output.wav gain -n -16

Step 5: Audio Effects

# Reverb: reverb [reverberance% HF-damping% room-scale% stereo-depth% pre-delay-ms wet-gain-dB]
sox input.wav output.wav reverb 50 50 100 100 0 0

# Echo / multiple echoes
sox input.wav output.wav echo 0.8 0.88 60 0.4
sox input.wav output.wav echos 0.8 0.7 700 0.25 700 0.3

# Equalizer (boost/cut frequencies)
sox input.wav output.wav equalizer 100 2q +6dB     # Boost bass at 100Hz
sox input.wav output.wav equalizer 3000 1q -4dB    # Cut mids at 3kHz

# Filters
sox input.wav output.wav highpass 80               # Remove rumble below 80Hz
sox input.wav output.wav lowpass 8000              # Remove hiss above 8kHz
sox input.wav output.wav bandpass 1000 200         # Center 1kHz, width 200Hz

# Noise reduction (two-step: profile then apply)
sox input.wav -n noiseprof noise.prof trim 0 0.5
sox input.wav output.wav noisered noise.prof 0.21

# Modulation effects
sox input.wav output.wav chorus 0.7 0.9 55 0.4 0.25 2 -t
sox input.wav output.wav flanger
sox input.wav output.wav tremolo 5 60
sox input.wav output.wav phaser 0.8 0.74 3 0.4 0.5 -t

# Speed (changes pitch) / Tempo (preserves pitch) / Pitch (preserves tempo)
sox input.wav output.wav speed 1.25
sox input.wav output.wav tempo 1.25
sox input.wav output.wav pitch 300     # Shift up 300 cents (3 semitones)

# Fade in/out: fade type in-length [stop-position out-length]
# types: t=linear, q=quarter-sine, h=half-sine, l=logarithmic, p=inverted-parabola
sox input.wav output.wav fade t 3 0 5

# Reverse
sox input.wav output.wav reverse

Step 6: Mixing & Combining

# Concatenate files (one after another)
sox file1.wav file2.wav file3.wav combined.wav

# Mix (overlay/combine simultaneously)
sox -m track1.wav track2.wav mixed.wav

# Mix with different volumes
sox -m -v 0.8 vocals.wav -v 0.3 music.wav mixed.wav

# Mix multiple tracks with individual levels
sox -m -v 1.0 vocals.wav -v 0.25 bgmusic.wav -v 0.15 sfx.wav final.wav

# Splice (insert one audio into another at a specific point)
# Use trim + concat approach:
sox original.wav part1.wav trim 0 30        # First 30s
sox original.wav part2.wav trim 30          # After 30s
sox part1.wav insert.wav part2.wav final.wav  # Concatenate with insert

# Crossfade between two files
sox file1.wav temp1.wav fade t 0 0 3        # 3s fade-out on first
sox file2.wav temp2.wav fade t 3 0 0        # 3s fade-in on second
sox -m temp1.wav temp2.wav crossfaded.wav   # Mix the overlapping parts

Step 7: Spectrograms & Visualization

# Generate spectrogram image
sox input.wav -n spectrogram -o spectrogram.png

# Customized spectrogram
sox input.wav -n spectrogram \
  -x 1200 -y 500 \           # Width x Height
  -z 80 \                     # Dynamic range (dB)
  -t "My Audio" \              # Title
  -c "Analysis" \              # Comment
  -o spectrogram.png

# Narrow spectrogram (specific frequency range)
sox input.wav -n spectrogram -x 800 -y 300 -z 90 -o spec.png rate 8000

# Stats output
sox input.wav -n stat
# RMS level, peak level, frequency info, etc.

sox input.wav -n stats
# More detailed: DC offset, crest factor, flat factor, peak count

Step 8: Batch Processing

# Convert all WAV to MP3
for f in *.wav; do sox "$f" "${f%.wav}.mp3"; done

# Normalize all files in a directory
for f in raw/*.wav; do sox "$f" normalized/"$(basename "$f")" norm -1; done

# Apply effects chain to all files
for f in episodes/*.wav; do
  sox "$f" processed/"$(basename "$f")" \
    highpass 80 norm -1 compand 0.3,1 6:-70,-60,-20 -5 -90 0.2 fade t 0.5 0 1
done

# Batch resample to 16kHz mono (for ML/speech)
for f in *.wav; do sox "$f" -r 16000 -c 1 resampled/"$(basename "$f")"; done

# Generate test tones
sox -n test_tone.wav synth 5 sine 440       # 5s 440Hz sine wave
sox -n pink_noise.wav synth 10 pinknoise    # 10s pink noise

Step 9: Effects Chains & Pipelines

# Podcast processing pipeline (chain multiple effects in one command)
sox raw_episode.wav final_episode.wav \
  highpass 80 noisered noise.prof 0.2 \
  compand 0.3,1 6:-70,-60,-20 -5 -90 0.2 \
  equalizer 3000 1q +2dB norm -1 fade t 1 0 2

# Pipe between sox instances (streaming)
sox input.wav -t wav - trim 10 60 | sox -t wav - output.wav norm -1

# Use with ffmpeg
ffmpeg -i video.mp4 -vn -f wav - | sox -t wav - processed.wav norm -1 highpass 80

Examples

Example 1: Process raw podcast recordings for publication

User prompt: "I have 12 raw podcast episodes in ./raw/ as WAV files. Remove low-frequency rumble, reduce background noise, compress the dynamic range, normalize to -1dB, and add a 1-second fade-in and 2-second fade-out to each."

The agent will:

1. Verify sox is installed with MP3 format support (

   sox --version

   and check for

   libsox-fmt-all

   ).
2. Profile background noise from the first 0.5 seconds of silence in the first episode using

   sox input.wav -n noiseprof noise.prof trim 0 0.5

   .
3. Create a

   ./processed/

   output directory.
4. Run a batch loop applying an effects chain:

   highpass 80

   ,

   noisered noise.prof 0.2

   ,

   compand 0.3,1 6:-70,-60,-20 -5 -90 0.2

   ,

   norm -1

   ,

   fade t 1 0 2

   to each file.
5. Report file sizes before and after processing.

Example 2: Prepare audio files for a speech recognition model

User prompt: "Convert all my interview recordings in ./interviews/ to 16kHz mono WAV files for Whisper transcription. Also trim silence from the start and end of each file."

The agent will:

1. Create a

   ./prepared/

   output directory.
2. Loop over all audio files in

   ./interviews/

   , running

   sox "$f" -r 16000 -c 1 prepared/"$(basename "$f")" silence 1 0.1 0.1% reverse silence 1 0.1 0.1% reverse

   to resample, convert to mono, and trim leading/trailing silence in one command.
3. Verify the output using

   soxi

   on a sample file to confirm 16kHz, mono, and reduced duration.

Guidelines

• Install

  libsox-fmt-all

  (or

  libsox-fmt-mp3

  ) on Debian/Ubuntu to enable MP3 and other format support; the base sox package only handles WAV and raw formats.
• Chain effects in a single sox command rather than piping between multiple sox processes; this avoids intermediate file I/O and preserves audio quality.
• Always use the two-step noise reduction workflow: profile a silent segment with

  noiseprof

  first, then apply with

  noisered

  at a sensitivity of 0.2-0.3 to avoid artifacts.
• When normalizing for podcasts, use

  norm -1

  (peak at -1dB) to leave headroom;

  norm -0

  risks clipping on some playback systems.
• Use

  soxi

  to inspect audio properties before processing; mismatched sample rates or channel counts between input files will produce unexpected mixing results.