Audio cheatsheet

Quick reference for common audio operations with sox and ffmpeg.
Both tools accept WAV, FLAC, MP3, OGG, and most other container/codec
combinations; pick whichever is already installed.

Trimming

Time format for both tools: HH:MM:SS[.ms], MM:SS[.ms], or seconds
(e.g. 90, 1:30, 0:01:30.250). Examples below trim from 00:10
for 30 seconds.

sox

sox takes trim <start> <length> after the output filename:

# Keep 30 seconds starting at 10 seconds.
sox in.wav out.wav trim 10 30

# Keep from 10 seconds through (absolute) 40 seconds — the leading
# '=' makes the second argument an end time rather than a duration.
sox in.wav out.wav trim 10 =40

# Drop a leading 2 seconds, keep the rest.
sox in.wav out.wav trim 2

# Trim silence from the start (anything below -50 dB for >0.1s).
sox in.wav out.wav silence 1 0.1 -50d

ffmpeg

-ss is the start, -t is the duration; -to is the absolute end.

# Re-encode (safest when accuracy matters):
ffmpeg -ss 00:00:10 -t 30 -i in.wav out.wav

# Stream-copy (fast, lossless; for audio-only inputs the cut is
# sample-accurate, so this is usually the right choice):
ffmpeg -ss 00:00:10 -t 30 -i in.wav -c copy out.wav

# Absolute end time instead of duration:
ffmpeg -ss 00:00:10 -to 00:00:40 -i in.wav -c copy out.wav

# Tip: putting -ss BEFORE -i is fast (seeks in the container); putting
# it AFTER -i is slower but more accurate for some video containers.
# For pure audio they behave the same.

Downsampling

"Downsampling" usually means one or more of:

• Sample rate (e.g. 44.1 kHz → 16 kHz)
• Bit depth (24-bit → 16-bit)
• Channel count (stereo → mono)

A common speech-friendly target is 16 kHz / 16-bit / mono — small
files that still capture the voice band.

sox

sox infers conversions from the output flags placed between the
input and output filenames.

# Sample-rate downsample only (sox uses a high-quality sinc filter by
# default, so there's no separate anti-aliasing step needed).
sox in.wav -r 16000 out.wav

# Bit-depth downsample only.
sox in.wav -b 16 out.wav

# Stereo to mono (average the channels).
sox in.wav -c 1 out.wav

# All three at once — the typical "downsample for ASR/transcription":
sox in.wav -r 16000 -b 16 -c 1 out.wav

# Mix only the left channel into a mono output (instead of averaging):
sox in.wav -c 1 out.wav remix 1

# Disable sox's automatic dither when reducing bit depth:
sox -D in.wav -b 16 out.wav

ffmpeg

# Sample-rate downsample only.
ffmpeg -i in.wav -ar 16000 out.wav

# Bit depth: WAV uses sample_fmt, e.g. s16 (16-bit signed PCM).
ffmpeg -i in.wav -sample_fmt s16 out.wav

# Stereo to mono.
ffmpeg -i in.wav -ac 1 out.wav

# All three at once:
ffmpeg -i in.wav -ar 16000 -ac 1 -sample_fmt s16 out.wav

# Lossy compressed target (Opus, suitable for voice at low bitrates):
ffmpeg -i in.wav -ar 16000 -ac 1 -c:a libopus -b:a 24k out.opus

Extracting audio from a video

Pulling the audio out of a video container is mostly an ffmpeg job.
The relevant flags:

• -vn — drop the video stream.
• -map 0:a:0 — select the first audio stream (useful when the
  source has multiple language/commentary tracks).
• -af "pan=mono|c0=c0" — keep only the first channel of that
  stream (e.g. just the left channel of a stereo recording), without
  averaging it with the others.
• -ac 1 — downmix all source channels to mono by averaging
  (different from selecting the first channel).

# Whole first audio stream → WAV, original channel layout preserved.
ffmpeg -i in.mp4 -vn -map 0:a:0 out.wav

# Just the FIRST audio channel of the first stream — copies channel 0
# only, no averaging.
ffmpeg -i in.mp4 -vn -map 0:a:0 -af "pan=mono|c0=c0" out.wav

# Same, plus speech-friendly downsample (16 kHz, 16-bit, mono).
ffmpeg -i in.mp4 -vn -map 0:a:0 -af "pan=mono|c0=c0" \
       -ar 16000 -sample_fmt s16 out.wav

# Alternative: downmix L+R to mono (averaged) instead of picking a
# single channel. Useful when both channels carry signal.
ffmpeg -i in.mp4 -vn -map 0:a:0 -ac 1 -ar 16000 -sample_fmt s16 out.wav

# Multi-stream sources: list available audio streams first.
ffmpeg -i in.mp4 2>&1 | grep "Audio:"
# Then -map 0:a:1 picks the second audio stream, etc.

Combining trim and downsample

Stack the operations in one command. The output is trimmed and
downsampled in a single pass.

# sox: trim 30s starting at 0:10, then downsample to 16 kHz mono 16-bit.
sox in.wav -r 16000 -b 16 -c 1 out.wav trim 10 30

# ffmpeg: same idea.
ffmpeg -ss 10 -t 30 -i in.wav -ar 16000 -ac 1 -sample_fmt s16 out.wav

Quick reference

Batch processing

# sox: convert every WAV in a directory to 16 kHz mono.
mkdir -p downsampled
for f in *.wav; do
  sox "$f" -r 16000 -c 1 "downsampled/$f"
done

# ffmpeg: parallelize with xargs (4 jobs at a time).
mkdir -p downsampled
ls *.wav | xargs -P 4 -I{} ffmpeg -i {} -ar 16000 -ac 1 -sample_fmt s16 downsampled/{}