Audio

Making and shaping sound in the graph: sources, filters, envelopes and sequencers running on their own clock, out to the audio device.

Sound runs on a clock of its own. The visible graph cooks once a frame; audio needs tens of thousands of samples a second, so the audio nodes are lifted out of the frame-rate graph into a small chain that a dedicated audio thread runs continuously. Numbers, signals & audio covers what a signal is and how it differs from a plain list of numbers.

Two things have to be true before anything is heard.

• The graph holds an Audio Output node, or an Audio Input one. Without at least one of the two, no chain is lifted and every audio node stays at frame rate.

• The timeline is playing. The audio thread is switched on at the play transition and off again at the pause, so a paused project is silent whatever is wired.

Sources

Audio Input takes a microphone or an interface, and Audio File plays a file from disk with its own start, end, speed and loop_mode, trimmed by dragging the in and out handles on its scrub bar. A rising edge on Audio File's trigger restarts it from the trim start, which is how a clip is fired from a step, a beat or a key. Its waveform output now reports the samples the speakers received rather than a separate cook-side estimate, so a trigger, a changed speed and the loop mode reach an analysis chain and the sound at the same time. Both nodes are covered in Inputs.

For synthesis, Oscillate is the one generator that reaches the audio thread. It makes sine, triangle, sawtooth, square and pulse waves plus white, pink and brown noise. Its sample_rate decides which clock it lives on: left at 0 it is a control wobble at frame rate, and raised it becomes an audible tone.

Shaping a signal

• Parametric EQ is three sweepable bands plus a low and a high cut. Each band has bandN_freq, bandN_gain in dB and bandN_q, defaulting to 100 Hz, 1000 Hz and 8000 Hz, and low_cut and high_cut bracket the lot. See Parametric EQ.

• Delay is an echo line. time is in milliseconds, 250 by default, feedback sets how much comes back round, and mix balances dry against wet. mode picks Simple, Ping Pong, which alternates across the stereo field, or Tape, which degrades each repeat. It writes a second channel on output_right.

• Dynamics compresses, limits, gates or expands, chosen on mode. threshold is in dB and defaults to -12, with ratio, attack and release in milliseconds, a soft knee and makeup_gain. Wire another signal into sidechain to duck one sound under another, and read gain_reduction to see how hard it is working.

• Filter is a biquad, switchable between Low-pass, High-pass, Band-pass and Notch with cutoff and resonance. It smooths a control value as readily as it filters a sound.

All four run on either clock. Give them audio and they process it sample by sample; give them a plain number and they do the same job once a frame, which is how a delay becomes a trailing animation and a compressor becomes a limiter on a slider.

Following a sound

• Pitch Detect reports the fundamental as frequency, midi_note and a readable note_name, with a confidence figure to gate on. min_freq and max_freq bracket the search, 50 Hz to 2000 Hz by default, and smoothing steadies the reading.

• Spectrogram keeps a scrolling history of the spectrum, history_frames deep, on a Linear or Log scale down to a db_floor of -60 dB. Its data output is a flat list of width by height numbers rather than a picture, so wire it into something that draws numbers.

• Envelope Follower and Audio Analysis turn sound into control values, from a single amplitude up to kick, snare, hi-hat and an estimated tempo. Both are covered in Inputs.

Sequencing and envelopes

• Step Sequencer runs steps steps, 16 by default, at a division of 1/8. direction covers Forward, Reverse, Pendulum, Random and Drunk, swing pushes the off-beats late, and gate_mode switches from a hand-drawn gate_pattern to Euclidean, which spreads euclidean_fills hits evenly across the bar and rotates them. Four value lanes, values_a to values_d, run alongside the gate, with probability and ratchet patterns per step.

• Piano Roll plays four voices from parallel lists of starts, pitches, velocities and durations, and emits note_1 to note_4 with matching velocities and gates, plus any_gate, playhead and phase. loop_mode is Loop, Once or Ping-Pong.

• ADSR shapes a level from a trigger. trigger fires when it crosses threshold on a Rising, Falling or Both edge, retrigger decides what a second hit does while the first is still running, and attack_curve, decay_curve and release_curve bend each stage away from a straight line.

• Value Tracker is not a sound node despite sitting in the same category. It accumulates a number over time as a Sum, Max, Min, Average, EMA, Rate or Count, with reset and gate inputs, an optional sliding window and optional bounds that Clamp or Wrap.

Both sequencers carry their own bpm, at 120. Setting Piano Roll's to 0 makes it follow the project tempo from the Score node; Step Sequencer has no such fallback and always uses the number in the box. See Tempo sync.

Reaching the speakers

Audio Output is the end of the chain. Wire a signal into left, and into right as well for stereo, and set volume and mute, both applied at the device. It reports the current level and the device's real device_sample_rate.

A frame-rate value wired into an audio node's control reaches it: the value is read once per cook and ramped smoothly across the samples in between, which is how a Step Sequencer drives an oscillator's frequency and how a slider changes a filter's cutoff without clicking.

Not connected yet

Five nodes here are complete and cannot be heard.

Audio Mixer never joins the audio chain, and because the chain is traced backwards from the output, it stops the trace at itself. Everything upstream of a Mixer therefore stays at frame rate as well, so a graph routed through one is silent rather than merely unmixed. Its master_gain, gain_N, mute_N and solo_N controls have no reader anywhere, and there are no pan controls at all, though the node describes them. There is no working way to sum two audio signals at audio rate today, and none to multiply them either: the Math node's audio side reads input ports the node does not have, so it writes silence.

Additive Synth is registered under a name the node no longer uses, so it is never lifted onto the audio thread and cannot produce sound. Wavetable and Granular are lifted, but the sample data they need never reaches the audio thread, so both render silence. All three still compute a correct buffer at frame rate, which can be read, drawn or analysed; none of it can be heard. The same gap silences Convolve used as a reverb, though its image side is unaffected. See Image effects.

Rendering audio to a file does not work either, for a separate reason. See Audio export.

See also

• Inputs

• Outputs

• Numbers, signals & audio

• Tempo sync