Decent Sampler compatibility

Voxo plays Decent Sampler presets and libraries — the .dspreset XML and the .dslibrary archive — from the same MIDI bytes the visuals draw. When a preset asks for something this version plays without, the load does not fail and nothing plays wrong in silence: the shell shows the compat report, once per load, calm, in the sentences below. They are the library’s own (voxo_note_copy); voxo/COMPAT_REPORT.md is their copy in the tree and the headless suite asserts the two never drift.

The first line of every report is the preset’s summary:

<name>: N zones in G groups, S samples, M MB in memory.

Then, in this order, only the lines that apply:

the note	the sentence
memory	Memory: this library is larger than advised for this device; it is loaded anyway.
missing samples	Samples that could not be read: those notes stay silent.
streaming	Disk streaming: this preset asks for it; everything is loaded to memory instead.
convolution	Convolution reverb: this preset uses it; it will play without it.
an unknown effect	An effect this version does not know: it will play without it.
sequences	Note sequences: this preset uses them; it will play without them.
unknown bindings	Bindings this version does not know: they are ignored.
the custom UI	Custom interface: not shown here; its controls’ starting values apply.

The memory line adds the decoded size against the shell’s advice (the desktop’s free memory with headroom); the missing-samples line adds the count and the first path with its reason; the unknown-effect and unknown-binding lines add the names in parentheses. A file that is not a preset at all is refused, and the text says why: not well-formed XML (with pugixml’s description and the byte), no <DecentSampler> root, an empty file, a file or an archive that cannot be opened, an archive holding no preset.

What plays at 3.0

Voxo DS 2.0 — the format coverage’s second generation, Phase 15 of the v3.0 arc — read the format’s own developer guide before each mechanism and retired three of the report’s sentences; the eight above are what remains: the housekeeping lines and the four features genuinely out of scope.

The sustain pedal (Voxo 0.27.0). CC 64 is a position: between up and down the release rate interpolates (the half-pedal), a note released under the full pedal decays at Voxo’s own sustainDecay (absent, the pedal holds, bit for bit the first release’s pedal), and the pedal pressed over a sounding tail catches it at its level (re-pedalling). Pedal noises fire once per crossing as voices of their own — onLoCCN / onHiCCN on a sample or group, the CC number in the attribute name, pitchKeyTrack="0" playing them at the root; loCCN / hiCCN plays a zone only with that CC’s last value in range (the pedal-down resonance); silencedByTags with silencingMode fast (the default, a 5 ms fade) or normal (the release phase) and silencingDecay is the choke, run before the new layers start.

The chorus and the phaser (0.28.0). <effect type="chorus"> plays at the instrument level and at the group level (one instance a note, as the format has it) with mix, modDepth, modRate; <effect type="phaser"> on the bus with mix, modDepth, modRate, centerFrequency, feedback. A binding on FX_MIX, FX_MOD_DEPTH, FX_MOD_RATE, FX_CENTER_FREQUENCY or FX_FEEDBACK resolves to the effect it points at; enabled="false" bypasses any effect. The shell’s CC map gains the chorus’s mix, rate and depth (targets 1006–1008).

The filter family, the filter envelope, the EQ, the gain and the wave shaper (0.29.0). The voice’s filter is any of lowpass (lowpass_4pl its legacy name), lowpass_1pl, highpass, bandpass and, at the group level, notch — with the format’s frequency and resonance, which is the Q (the first release read it as an amount: every lowpass preset sits 3 dB brighter at its cutoff since). peak, low_shelf, high_shelf and an instrument-level notch are the bus EQ, in their listed order, eight stages at most; gain plays in dB ("-6dB" allowed, levelUnit="linear"); wave_shaper plays at the bus and the group level with drive, driveBoost, outputLevel. An <envelope> bound to a filter’s frequency or Q is the group’s filter envelope.

The modulator matrix, legato and glide (0.30.0). <lfo> plays — shape sine, triangle, square or saw; frequency in hertz or, under frequencyFormat="musical_time", the guide’s subdivision index on the session’s tempo; delayTime, modAmount, scope global or voice, modBehavior set, add, multiply or modulate — and <envelope> with its delay and hold stages beside the ADSR. Every binding under either is a target: the volume, the pan, the group and global tuning, the voice filter’s frequency and Q, a group effect’s parameters, the bus effects’ and the EQ stages’ — eight modulators of eight targets at most. glideTime and glideMode (legato, always, off), polyphony, trigger="first" / "legato", previousNotes and legatoInterval play as the guide documents them.

The instrument slot (0.31.0). A session preset and a recording’s header carry the library the sound played — by name and by the hash of its preset XML, with the path as a hint, never embedded; each shell resolves it against its own libraries and says when none is there. The offline bounce renders a recording to a 32-bit float WAV through the same engine, deterministically; the output device is chosen in the desktop’s Sound settings.

Out of scope

Still not played, and said so in the report: disk streaming (everything is loaded to memory), convolution reverb, the note sequences and the custom interface. Also not played, silently: trigger="continuous", releaseTriggerDecay, tag polyphony limits, and a group-level peak or shelf (outside the format). The record of each decision is the decision log’s Part XIV (DECISIONS_14 #1–#5); the Sound guide says how to load a library on each platform.