SoundStudio Audio Theory

Audio theory tutorial

This page explains the ideas behind every tool in SoundStudio, from what a sound wave is to why a song gets louder without clipping. Each section ends with something to try in the editor. Terms in the text are defined in the Glossary. Read it top to bottom once; after that, jump to the section you need.

1. What sound is

Sound is air pressure going up and down very fast. A speaker cone pushes air forward and pulls it back; your eardrum follows the same motion. Draw that pressure over time and you get a waveform.

A pure tone. Height is amplitude (loudness); the length of one repeat is the period. The number of repeats per second is the frequency.

Frequency is how many cycles happen each second, in hertz (Hz). 100 Hz is a low hum; 1,000 Hz (1 kHz) is a whistle-like tone; 10 kHz is a thin sizzle. People hear roughly 20 Hz to 20 kHz, and the top end falls with age, often to 14 or 15 kHz by middle age.
Pitch is what we hear from frequency. Doubling the frequency raises the pitch one octave: A below middle C is 220 Hz, the A above it 440 Hz, the next 880 Hz. An octave holds 12 semitones, each about 6% higher than the last.
Amplitude is how big the swings are. Bigger swings are louder.
Harmonics: real instruments and voices are not pure tones. A sung A at 220 Hz also contains 440, 660, 880 Hz and so on, each at its own level. That recipe of harmonics is why a guitar and a voice on the same note sound different (timbre).

Try it: Tools tab, Generate, Tone makes a 440 Hz pure tone. Zoom in until you see the smooth wave. Then load any song and zoom in the same way: the wave is far more jagged, because it is many frequencies added together.

2. Digital audio: samples, sample rate and bit depth

A computer cannot store a smooth curve, so it measures the height of the wave many times a second. Each measurement is a sample.

Sampling. The sample rate sets how often the height is measured; the bit depth sets how finely each height can be written down.

Sample rate

The number of samples per second. 44,100 (44.1 kHz) is the CD and music standard; 48,000 (48 kHz) is used for video and broadcast. A sample rate can only hold frequencies up to half its value, called the Nyquist frequency: 44.1 kHz audio holds up to 22.05 kHz, just past the top of human hearing. Anything higher in the original has to be filtered out first, or it folds back down as harsh false tones (aliasing).

Bit depth

How many bits describe each sample. Each bit adds about 6 dB of range between the loudest sound and the finest detail:

Bit depth	Range	Use
16-bit	about 96 dB	CDs and final delivery. Plenty for listening.
24-bit	about 144 dB	Recording and editing. The noise floor is far below anything you will hear, so repeated edits do not build up grain.
32-bit float	huge	What editors use inside. It can even go past full scale for a moment without damage, as long as it is brought back down before export.

SoundStudio works in 32-bit float internally and exports 16 or 24-bit WAV. Keep a 24-bit copy of anything you may edit again.

Try it: zoom all the way in on any file. At the highest zoom you can see the individual samples as tiny steps.

3. Decibels

Our ears handle a huge range, from a pin drop to a jet engine, so audio uses a logarithmic scale: the decibel (dB). Decibels describe a ratio, not an amount. The rules worth memorizing:

+6 dB doubles the height of the waveform; -6 dB halves it.
About +10 dB sounds twice as loud to most people.
1 dB is about the smallest change most people notice in music.

In a digital file the top of the scale is 0 dBFS (decibels relative to full scale). Nothing can be stored louder, so every level is a negative number: a peak at -6 dBFS is half of full scale.

dBFS against waveform height. Most of what you hear lives between -30 and 0 dBFS.

The space between your loudest peak and 0 dBFS is headroom. Leave some. MP3 conversion, sample-rate changes and playback on other devices can all push peaks a little higher than they were.

Try it: open the Meters tab and play a song. The bar shows the peak level in dBFS. Then use Tools, Apply Gain at -6 dB and watch the peaks drop by exactly 6 while the song sounds a bit quieter, not half as loud.

4. Waveform and spectrum: two views of the same sound

The waveform shows level over time. A spectrum shows level by frequency: how much bass, mids and treble are in the sound right now. The math that converts one to the other is the FFT (fast Fourier transform). A spectrogram combines both: time across, frequency up, loudness as brightness.

Different problems are easier to spot in different views. A click is a sharp spike in the waveform and a thin vertical line in a spectrogram. Hum is invisible in the waveform but shows as a bright horizontal line at 60 Hz (50 Hz in Europe). Hiss shows as a haze across the top of the spectrogram.

How a click looks in each view. Music is made of lines that run across the spectrogram; a click is a line that runs up it.

Try it: the Meters tab shows a live spectrum while playing. In the Noise tab, a found click's report lists its frequency range, taken from this kind of analysis.

5. Loudness: peak, RMS and LUFS

Two songs can both peak at -1 dBFS and still sound very different in volume. Peaks only measure the single highest instant. How loud something sounds depends on its average level and on which frequencies carry the energy; ears are most sensitive around 2 to 5 kHz and much less sensitive to deep bass.

Measure	What it tells you
Peak	The highest instant. Tells you if it will clip, not how loud it sounds.
True peak (dBTP)	The peak between the samples, found by rebuilding the waveform in finer steps. Needed because a file can clip on playback even when no sample reaches 0 dBFS.
RMS	The average power over a short window. Closer to loudness, but treats all frequencies the same.
LUFS	The broadcast standard (ITU-R BS.1770). Filters the sound the way ears hear it, averages over the whole song and skips silences. This is what streaming services and broadcasters use.

Loudness targets

Where it plays	Typical target
Spotify, YouTube, Amazon Music	about -14 LUFS
Apple Music, podcasts	about -16 LUFS
TV in Europe (EBU R128)	-23 LUFS
TV in the US (ATSC A/85)	-24 LUFS
Peak ceiling for most delivery	-1 dBTP (some broadcasters ask for -2)

Streaming services turn loud songs down to their target. A song squashed to -8 LUFS gets turned down by 6 dB and just sounds flatter than a song mastered at -14 with its punch intact. That is why SoundStudio's Enhance stops at -1 dBTP instead of limiting harder to chase a number.

Try it: Enhance tab, Analyze only reports LUFS, true peak and loudness range without changing anything.

6. Clipping and distortion

When a signal tries to go past 0 dBFS, the file cannot store it, so the tops of the waves are cut flat. That flat edge adds new harmonics that were never in the music: clipping. A little is a crunch on drum hits; a lot is harsh, fuzzy and tiring.

Clipping flattens every peak that crosses the ceiling.

Distortion is not always bad. Guitar amps, tape and tube gear add pleasant, gentle distortion on purpose. The difference is control: soft rounding sounds warm; hard flat tops sound broken.

The cure is prevention: leave headroom and turn things down evenly rather than letting them hit the ceiling. Once a peak is clipped, repair can only guess the missing top. SoundStudio's Repair Clipping predicts it from the audio on both sides, which works well on short flat spots of a few milliseconds.

Try it: Tools tab, Detect Clipping. If it finds flat runs, Repair Clipping rebuilds them, and Ctrl+Z puts them back if you prefer the original.

7. Clicks, pops and ticks

A click is a burst of sound far shorter and sharper than anything musical, usually a fraction of a millisecond to a few milliseconds. Common causes:

An edit that cuts the waveform mid-swing, so it jumps instantly from one level to another.
Digital glitches: dropped samples, buffer problems while recording, USB and clock trouble.
AI generators such as Suno, which sometimes leave tiny broadband ticks in their output.
Vinyl dust and scratches; mouth clicks and lip smacks on voice recordings.

Why repair beats deleting

Deleting even 1 ms shifts everything after it earlier by 1 ms and leaves a seam that can click on its own. Repair instead rebuilds the damaged samples in place. SoundStudio's repair (LSAR) studies the audio on both sides of the click, learns its pattern of resonances, and fills the gap with the samples that best continue that pattern from both directions at once. On a sustained note the result is usually impossible to hear.

Deleting is still right for things that should not be there at all, like a stray breath before the first word, as long as the cut is blended (crossfaded).

Why some clicks are hard to find

A detector compares each moment with what the audio around it predicts. In a quiet passage a click stands out clearly. In a dense, distorted chorus the music itself is full of sharp edges, so a click can hide inside it. That is where your eyes and ears beat any detector, and where the Spike Pen helps: you point at the spike you see and it repairs just that one.

Try it: Noise tab, Find Clicks, then click a red mark and Fix This Click. Zoom in first and look at the spike, then again after: the waveform should continue smoothly through the repaired spot.

8. Noise: hiss, hum and rumble

Noise	Sounds like	Where it lives	Fix
Hiss	Air, tape noise, "ssss"	Mostly above 4 kHz, steady	Noise reduction with a profile, or AI noise reduction
Hum	Low steady buzz	60 Hz (North America) or 50 Hz, plus 120, 180 Hz...	Narrow cuts at the hum frequency and its multiples
Rumble	Thumps, traffic, handling	Below about 80 Hz	High-pass filter (low cut)
Room tone, chatter	Background of the space	Everywhere	AI noise reduction; better, a quieter room

Profile-based noise reduction listens to a stretch of noise alone, learns its spectrum, then subtracts that much of each frequency from the whole file. Push it too far and you get musical noise, a watery warble, because the noise is random and cannot be removed exactly. AI noise reduction (DeepFilterNet in the Helper) has learned what voices sound like and keeps them, which works much better on speech, but it can treat music under a voice as noise.

Rule of thumb: reduce noise until it stops being distracting, not until it is gone. A little natural background sounds more real than a perfectly silent, processed gap.

Try it: select half a second of noise only, Noise tab, Capture Noise Profile, then Apply at 50 to 70%. Compare with Ctrl+Z and Ctrl+Y.

9. EQ: shaping tone

An equalizer turns frequency ranges up or down. Knowing roughly what lives where makes EQ much faster than guessing:

The audible range on a log scale, with what each region does.

Filter types: a high-pass (low cut) removes everything below a point; a shelf raises or lowers everything above or below a point; a bell (peaking) band raises or lowers a region around one frequency. Q sets how wide a bell is: low Q broad and musical, high Q narrow and surgical.
Cut before you boost. Removing mud at 250 to 400 Hz often makes a voice clearer than adding treble.
Small moves. 2 to 3 dB is a big change. Over 6 dB usually means another problem (the microphone, the room).
Compare at equal loudness. Louder always sounds "better" for a moment, so match levels before judging.

Try it: EQ tab. Set Low Cut to 80 Hz on a voice recording and listen for handling noise and rumble disappearing while the voice stays full.

10. Dynamics: compressors, limiters and gates

A compressor turns the sound down automatically whenever it gets louder than a threshold. The ratio says how much: at 4:1, a sound 8 dB over the threshold comes out only 2 dB over. Attack is how fast it clamps down, release how fast it lets go, knee how gently it starts. Afterwards makeup gain brings the whole thing back up, so quiet parts end up relatively louder and the voice sits steadier.

A compressor's transfer curve. Below the threshold nothing changes; above it, output rises more slowly than input.

Voice: ratio 2:1 to 4:1, attack 5 to 20 ms, release 100 to 250 ms, 3 to 6 dB of reduction on the loud words.
Limiter: a compressor with a very high ratio and a fast attack; a safety ceiling for the last dB or two, not a way to make things loud.
Gate / expander: the opposite. It turns quiet parts down further, to hush noise between phrases. Set gently, or word endings get chopped.

Too much compression removes the life from music: everything is equally loud, so nothing hits. If you can clearly hear the compressor working, back it off.

Try it: Dynamics tab. Start at threshold -20 dB, ratio 3:1, and raise the threshold until only the loudest words are affected.

11. Time and pitch

On tape, playing faster makes the sound both shorter and higher, because every wave is squeezed. Digital audio can separate the two:

Resampling (tape style) changes both; time stretch changes only length; pitch shift changes only pitch.

Time stretch (and SoundStudio's Fit to Length) cuts the audio into short overlapping pieces, about 30 ms each, and overlaps them a bit closer together (shorter) or further apart (longer). Each piece is nudged so its waveform lines up with the one before, which avoids flanging and keeps the pitch. This method is called WSOLA.
Pitch shift stretches the audio longer or shorter, then resamples it back to the original length, which moves the pitch.
Formants: shifting a voice up also shifts the resonances of the throat and mouth, which causes the chipmunk sound. Better tools, like Rubber Band in the Helper, keep formants in place.

How far can you go? Up to about 10% faster is very hard to hear on speech. 10 to 20% works for most reads, especially if the pauses are shortened first. Past about 20% it starts to sound rushed. Music is less forgiving than speech; a few percent is usually safe.

Try it: Effects tab, Time Stretch at 90% on a song, then Pitch Shift +2: the first slows it down at the same pitch, the second raises the key at the same length.

12. Stereo, phase and mono

Stereo uses two channels. A sound equally loud in both seems to come from the center (that is where lead vocals, kick and bass usually sit); louder in one side pulls it toward that side (panning). Phase matters when channels are combined: if the left channel is the exact upside-down copy of the right, adding them to mono gives silence. Always check that a mix still sounds right in mono, because phones, smart speakers and many radios play mono.

Center-panned vocals are also why simple voice isolation works: subtracting left from right cancels what is identical in both channels. It is crude because it also removes bass and drums in the center; AI stem separation (Demucs in the Helper) does the job properly.

Try it: Tools tab, Convert to Mono, and listen for anything that gets quieter or hollow. Ctrl+Z to go back.

13. Clean edits

Every cut joins two points of the waveform. If one side is high and the other low, the waveform jumps instantly, and an instant jump is a click. Two defenses:

Zero crossings: cut where the waveform crosses the center line, so both sides start near zero.
Crossfades: blend the end of one side into the start of the other over a few milliseconds. SoundStudio does this automatically on every cut and paste.

On a sustained note, even a crossfaded cut can be heard, because the note's waveform is in a different part of its cycle on each side. Smooth Cut moves the end point by a few milliseconds to where the two sides line up, then blends over 20 ms. That is how to shorten a held note without a bump.

Try it: select part of a long sung note, then Tools tab, Smooth Cut (Shift+Delete). Compare with a plain Delete on the same spot.

14. File formats

Format	Type	Size (stereo, per minute)	Use for
WAV 24-bit, 48 kHz	Uncompressed	about 17 MB	Editing, masters, archive
WAV 16-bit, 44.1 kHz	Uncompressed	about 10 MB	CD-quality delivery
FLAC	Lossless	roughly half of WAV	Archive and sharing at full quality
MP3 320 kbps	Lossy	about 2.4 MB	Sharing, uploads, players
AAC / M4A 256 kbps	Lossy	about 1.9 MB	Apple devices, video
Opus 160 kbps	Lossy	about 1.2 MB	Web, messaging, small files

Lossy formats lose a little each time they are re-encoded. Editing an MP3 and saving it as MP3 again stacks the damage, especially on cymbals and reverb tails. Work in WAV (or FLAC), and make the MP3 last, once. If your source is an MP3, as Suno downloads often are, export your edited version as 24-bit WAV and only make an MP3 at the very end.

15. Putting it together: cleaning up an AI-generated song

Download the best version the service offers (WAV if possible) and open it. Autosave has your back from here.
Listen first, all the way through, with headphones. Note the times of anything that bothers you.
Clicks: Noise tab, Find Clicks with the Detector set to Suno ticks, then again with Spikes above the outline. Review the strongest first, Fix This Click on the real ones, and use the Spike Pen for ones you can see but the detector missed.
Hard spots: if a click is buried in a busy chorus, separate stems (Helper tab), fix it on the vocal stem where it stands out, then Mix Files Together.
Timing and length: shorten an overlong held note with Smooth Cut; remove a bar or a repeat with a plain Delete at a beat, crossfaded.
Tone: only if needed. A gentle low cut at 30 Hz and small EQ moves; AI songs are usually already mixed.
Loudness: Enhance tab, target -14 LUFS, -1 dBTP ceiling. Check the report.
Export a 24-bit WAV master, then an MP3 320 or FLAC for sharing.

16. Putting it together: a radio spot

Record or load the voice read. Trim silence at the head and tail (Noise tab, Silence Removal, Trim Start/End).
Clean it: low cut at 80 Hz, noise reduction if the room was noisy, remove mouth clicks with the click tools.
Compress lightly so every word is clear: about 3:1, 3 to 6 dB of reduction.
Fit to Length: type 30 (or 15, 60) and choose Tighten lightly first. Check that the speed-up stays under about 15%; if not, trim a phrase or re-read.
If there is a music bed, fit the voice first, then mix it over a bed of the right length, with the music about 15 to 20 dB under the voice.
Loudness: follow the station's spec. Commonly -24 LUFS with -2 dBTP for US broadcast, -23 LUFS for European broadcast, -16 LUFS for podcast or streaming ads.
Export in the format the station asks for, usually 48 kHz or 44.1 kHz WAV, 16 or 24-bit.