scale

Audio scaling utility

The scale utility scales the amplitude of sound files by constant factors, time-varying curves, or normalizes to a target level. It can also analyze and report the maximum possible scaling without clipping.

Syntax

scale [-flags] soundfile

Output options

Option	Description
-o <filename>	Output sound filename (default: test)
-A	Create AIFF format output
-W	Create WAV format output
-J	Create IRCAM format output
-h	No header (raw output)

Output sample format

Option	Description
-c	8-bit signed char
-a	A-law encoded
-u	μ-law encoded
-s	16-bit short int (default)
-l	32-bit long int
-f	32-bit float

Scaling options

Choose one of these methods:

Option	Description
-F <factor>	Fixed scale factor
-F <filename>	Time-varying scale curve file
-M <maximum>	Scale to given maximum amplitude
-P <percent>	Scale to percentage of full scale

Additional options

Option	Description
-R	Continuously rewrite header during writing
-H <n>	Heartbeat: 1=spinner, 2=dots, 3=numeric
-N	Ring bell when complete
-- <file>	Log output to file

Examples

Fixed amplitude scaling

Reduce by 6 dB (50%):

scale -F 0.5 input.wav -o output.wav

Double amplitude (+6 dB):

scale -F 2.0 input.wav -o output.wav

Reduce by 20 dB (10%):

scale -F 0.1 input.wav -o quiet.wav

Normalize to maximum level

Normalize to 100% full scale:

scale -M 1.0 input.wav -o normalized.wav

Normalize to -3 dB (0.707 full scale):

scale -M 0.707 input.wav -o output.wav

Normalize to -6 dB (0.5 full scale):

scale -M 0.5 input.wav -o output.wav

Scale to percentage

Scale to 75% full scale:

scale -P 75 input.wav -o output.wav

Scale to 50% (same as -M 0.5):

scale -P 50 input.wav -o output.wav

Find maximum possible scaling

scale -F 0.0 input.wav

Output:

Maximum scaling without clipping: 2.456
Peak amplitude: 0.407

Use this to determine safe scale factor:

scale -F 2.4 input.wav -o output.wav

Time-varying amplitude curve

scale -F envelope.txt input.wav -o output.wav

With envelope.txt containing:

0 0.0
44100 1.0
220500 1.0
264600 0.0

Format: sample_number scale_factor

This creates a fade in (0-1s) and fade out (5-6s) at 44.1kHz.

Scaling methods

Fixed factor (-F)

• Multiplies all samples by constant value
• factor = 0.5 reduces amplitude by 6 dB
• factor = 2.0 increases amplitude by 6 dB
• Linear scaling: output = input × factor

Maximum amplitude (-M)

• Analyzes peak level in input
• Calculates factor to reach target maximum
• Formula: factor = target / current_peak
• Preserves relative dynamics

Percentage (-P)

• Similar to -M but specified as percentage
• -P 100 = full scale (same as -M 1.0)
• -P 50 = half scale (same as -M 0.5)
• -P 75 = 75% of full scale (same as -M 0.75)

Time-varying curve (-F filename)

• Reads breakpoint file with sample/gain pairs
• Linear interpolation between breakpoints
• Allows complex amplitude envelopes
• Good for fades, dynamics, automation

Amplitude and decibels

Conversion between linear and dB:

Factor	dB	Effect
0.1	-20 dB	Very quiet
0.25	-12 dB	Quiet
0.5	-6 dB	Half amplitude
0.707	-3 dB	Half power
1.0	0 dB	No change
1.414	+3 dB	Double power
2.0	+6 dB	Double amplitude
4.0	+12 dB	Loud
10.0	+20 dB	Very loud

Formulas:

• dB to factor: factor = 10^(dB/20)
• Factor to dB: dB = 20 × log10(factor)

Normalization strategies

Peak normalization

Most common, using -M or -P:

scale -M 1.0 input.wav -o normalized.wav

Pros:

• Maximizes level without clipping
• Simple and predictable
• Preserves dynamics

Cons:

• Peak may be isolated transient
• Average level may still be low
• Different perceived loudness across files

Conservative normalization

Leave headroom for safety:

scale -M 0.9 input.wav -o output.wav  # -1 dB
scale -M 0.8 input.wav -o output.wav  # -2 dB
scale -M 0.707 input.wav -o output.wav  # -3 dB

Use when:

• Further processing planned
• Preventing inter-sample peaks
• Working with lossy codecs

Batch normalization

Normalize multiple files to same level:

for file in *.wav; do
  scale -M 0.9 "$file" -o "normalized/$file"
done

Creating scale curves

Manual curve creation

Linear fade in (0 to 1 over 2 seconds at 44.1kHz):

0 0.0
88200 1.0

Linear fade out (1 to 0 over 3 seconds at 48kHz):

0 1.0
144000 0.0

Crossfade shape:

0 1.0
44100 0.0

For second file (inverse):

0 0.0
44100 1.0

Generated curves

Python script for exponential fade:

import math

srate = 44100
duration = 2.0
samples = int(srate * duration)

# Exponential fade out
for i in range(0, samples, 100):
    # Exponential decay
    t = i / samples
    gain = math.exp(-3 * t)  # -3 gives smooth curve
    print(f"{i} {gain}")

Python script for S-curve fade:

import math

srate = 44100
duration = 1.0
samples = int(srate * duration)

# S-curve (smooth acceleration/deceleration)
for i in range(0, samples, 100):
    t = i / samples
    # Smoothstep function
    gain = t * t * (3 - 2 * t)
    print(f"{i} {gain}")

Common use cases

Prepare for mastering

scale -M 0.8 mix.wav -o premaster.wav

Leaves 2 dB headroom for mastering processing.

Match levels across files

Find peak levels:

scale -F 0.0 file1.wav
scale -F 0.0 file2.wav

Normalize to same target:

scale -M 0.7 file1.wav -o file1_norm.wav
scale -M 0.7 file2.wav -o file2_norm.wav

Add fade in/out

Create envelope file fade.txt:

0 0.0
22050 1.0
529200 1.0
551250 0.0

Apply:

scale -F fade.txt input.wav -o faded.wav

Prevent clipping after mixing

# Check level after mixing
scale -F 0.0 mixed.wav
# Output: Maximum scaling: 0.83
# Peak exceeds full scale!

# Scale down
scale -F 0.8 mixed.wav -o fixed.wav

Technical notes

Floating point accuracy

Use float output to avoid quantization:

scale -f -F 0.5 input.wav -o output.wav

Benefits:

• No quantization noise
• Can exceed ±1.0 range
• Useful for intermediate processing

Clipping behavior

Integer formats clip at maximum value:

• 16-bit: ±32767
• 24-bit: ±8388607
• 32-bit: ±2147483647

Float formats don’t clip but may cause problems downstream.

Related utilities

• mixer - Mix files with scaling
• sndinfo - Check file properties
• srconv - Sample rate conversion