Modulation effects, advanced spatial effects and granular/spectral audio processing.

Modulation effects, advanced spatial effects and granular/spectral audio processing

Modern audio processing ranges from classic time and amplitude modulation effects to complex mathematical transformations in the frequency domain ($STFT$) and the domain of tiny time fragments (granular synthesis). Mastering this broad processor palette lets you radically transform static sounds into striking, three-dimensional experiences.

---

1. Time- and amplitude-domain modulation effects

Modulation effects use a low-frequency oscillator (LFO) to cyclically alter time, gain or phase parameters.

flowchart LR
    In["Input Audio"] --> Split["Signal Splitter"]
    Split --> Dry["Direct Signals (Dry)"]
    Split --> DelayLine["Variable Delay Line (Delay Line)"]
    LFO["LFO Oscillator (0.1 Hz - 10 Hz)"] --> DelayLine
    DelayLine --> Modulation["Delay Time / Pitch Adjustment"]
    Dry --> Mix["Summer (Dry + Wet)"]
    Modulation --> Mix
    Mix --> Out["Output Processed with Modulation"]

Chorus

• Mechanism: the signal is duplicated and sent through one or more LFO-modulated delay lines with times varying between $15\text{ ms}$ and $30\text{ ms}$.
• Result: slight pitch detuning and phase offsets simulate several performers playing in unison.

Tremolo and auto-pan

• Tremolo: periodically modulates signal amplitude with an LFO using sine, triangle or square waveforms. It creates rhythmic volume pulsation.
• Auto-pan: modulates stereo balance ($Panorama$), alternating level between left and right channels in opposite phase ($180^\circ$). Synced to track tempo ($1/4, 1/8$), it is ideal for adding movement to hi-hats and arpeggios in DJ mixes.

Ring modulation

• Mechanism: multiplies the input audio waveform point by point with an audible-frequency carrier wave ($100\text{ Hz} - 4\text{ kHz}$).
• Result: cancels the original fundamental and produces only sums and differences of spectral components ($f_{carrier} + f_{in}$ and $f_{carrier} - f_{in}$). It creates metallic, robotic, bell-like inharmonic sounds.

---

2. Spatial delays: ping-pong delay

Ping-pong delay is a stereo variation of traditional delay whose repeats alternate consecutively between left and right speakers.

• Routing architecture: the first processed repeat appears in the left channel; part of its feedback tail feeds the right-channel input, bouncing rhythmically between both panorama extremes.
• Tempo alignment: in the DJ booth, ping-pong delays synced to dotted eighth notes ($3/16$) or triplets ($1/3$) create enveloping width without covering the centred vocal in the Mid bus.

---

3. Convolution reverb

Algorithmic reverbs simulate spaces using complex delay and comb-filter networks, while convolution reverb captures the actual acoustic signature of existing physical rooms.

Impulse responses ($IR$)

To create an $IR$, a perfect impulse signal, such as a blank gunshot or a sine sweep, is emitted inside a cathedral, theatre or studio, and calibrated microphones record the complete reflection response.

Mathematical convolution operation

Discrete convolution between audio signal $x[n]$ and impulse response $h[n]$ is defined as: $$y[n] = (x * h)[n] = \sum_{k=0}^{M-1} x[k] \cdot h[n - k]$$

Because computation in the time domain is expensive, modern convolution processors transform audio into the frequency domain using the Fast Fourier Transform (FFT), directly multiply spectra and transform the signal back using the IFFT (overlap-add method).

---

4. Vocal processors: vocoder and harmonizer

Vocoder

A vocoder analyses the spectral envelope of a modulator signal, usually the human voice, and applies it to a carrier, a harmonically rich synthesiser such as a sawtooth pad.

1. Analysis filter bank: splits the voice into $16 - 32$ band-pass frequency bands.
2. Envelope followers: extract each vocal band’s amplitude curve in real time.
3. Synthesis filter bank: modulates the carrier synthesiser’s VCAs using the extracted envelopes.
4. Result: the synthesiser speaks or sings while retaining the musical notes played on a keyboard or MIDI sequencer.

flowchart TD
    Mod["Voice (Modulator)"] --> ModBank["Analysis Filter Bank"]
    ModBank --> EnvFollowers["Envelope Followers"]
    
    Carrier["Synthesiser (Carrier)"] --> CarrierBank["Synthesis Filter Bank"]
    EnvFollowers --> CarrierBank
    CarrierBank --> Out["Synthesised Vocal Output (Vocoder)"]

Harmonizer

A harmonizer analyses the incoming voice’s fundamental frequency ($f_0$) using autocorrelation detection or PSOLA (Pitch-Synchronous Overlap-Add) processing. It then generates copies tuned to specific harmonic intervals, such as major thirds, perfect fifths or octaves, building complete polyphonic backing vocals from one monophonic vocal line.

---

5. Bitcrusher: digital resolution reduction

A bitcrusher deliberately distorts digital audio by degrading two key quantisation parameters:

1. Bit-depth reduction: reduces the signal’s vertical resolution. Dropping from 24 bits to 4 bits makes quantisation steps enormous, generating loud, harsh quantisation noise.
2. Sample-rate reduction (downsampling): holds the sample value constant over several cycles ($Sample & Hold$), violating the Nyquist theorem and introducing severe aliasing distortion with a retro 8-bit console-audio texture.

---

6. Granular and spectral processing

Granular processing (granular FX)

Granular processing divides continuous audio into tiny fragments called grains, lasting between $1\text{ ms}$ and $100\text{ ms}$.

• Each grain receives an individual smooth window, such as Hanning or Gaussian, to avoid clicks.
• Grains can play at different speeds, backwards, with stochastic pitch variation, or scattered randomly across time and stereo space.
• Application: texture clouds, infinite pads, atmospheric DJ-set transitions and time freezing.

Spectral processing (spectral FX / FFT filtering)

Spectral processing splits audio into short time windows and transforms them into the frequency domain using the STFT (Short-Time Fourier Transform).

• It lets you manipulate the magnitude and phase of thousands of individual frequency bins independently.
• Advanced applications: surgical noise removal with a spectral noise gate, harmonic spectral freezing, AI stem demixing and gradual timbral morphing between two songs.

---

7. Summary of specialist and spatial processors

Processor	Working domain	Key parameter	Editing / DJ purpose
Chorus	Time ($15-30\text{ ms}$)	LFO Rate / Depth	Added stereo texture and thickness
Tremolo	Amplitude	LFO Shape / Depth	Rhythmic volume pulsation
Ping-Pong Delay	Time + Stereo	Feedback / Time Sync	Bouncing L/R-channel repeats
Convolution Reverb	Physical space ($IR$)	Loaded impulse	Ultra-realistic recreation of real rooms
Vocoder	Coupled spectrum	Bands / Modulator	Robotic vocals articulated by a synthesiser
Harmonizer	Pitch / Frequency	Musical Scale / Shift	Complex vocal harmonies
Bitcrusher	Digital quantisation	Bits / Downsample	Retro digital lo-fi distortion
Granular FX	Micro-windows ($ms$)	Grain Size / Density	Texture clouds and audio freezing
Spectral FX	Frequency (FFT)	FFT Bins / Window Size	Surgical spectral and stem editing