Slab 0.0.14

Thanks to Lost Robot, the author of septabee, for pointing out the
audio-thread inefficiencies that this release fixes.

• C-Beam 80, a Yamaha CS-80. Eight notes of two complete voices each.
  Each voice has a linear-in-Hz VCO, an HPF and an LPF that don't
  self-oscillate, and the IL/AL filter envelope behind CS-80 brass. Velocity
  is a fading initial-touch accent. Per-note pressure opens the filter,
  raises the level and brings in vibrato on the pressed notes only. The
  sub oscillator, ring modulator, BBD chorus/tremolo, keyboard scaling,
  constant-rate portamento and sustain are global, as on the original.
  It ships with Yamaha's factory pairings from the Owner's Manual charts
  and four sounds of ours. See docs/38.

• Choose the audio buffer size. The audio menu next to record lists
  64 to 1024 frames with their latency; 64 is 1.3 ms. The choice is
  remembered.

• Notes played at idle sound at once. A key pressed after the app
  had been quiet could wait up to 50 ms before reaching the engine.

• Recorded takes land in time. Placement now uses the latency the
  audio device reports each way, plus how long the input actually
  waited, instead of a two-buffer guess. With AirPods as the output,
  takes used to land about 150 ms late.

• Record from any input, mono or stereo. Right-click a track's R to
  pick its input: In 1, In 2, … or a stereo pair. Before, every input
  of an interface was mixed into one, 6 dB down and with the other
  inputs' noise. The choice is saved with the project.

• Recording holds up when things go wrong. Audio the disk couldn't
  keep up with becomes silence instead of shifting the rest of the take
  earlier; seeking or stopping ends the take where it still lines up; a
  loop pauses for the take. Unplugging a device, a sample-rate change
  from another app, or the system audio service restarting now reopens
  the device instead of leaving it silent.

• The Sampler no longer aliases when you play high. Clean playback
  now band-limits the sample to the speed it is read at, using octave
  copies made when the sample loads, so playing a sample an octave or
  more above its root, or a 96 kHz file, no longer folds whistles and
  grit back into the mix.

• No more Sampler clicks on retrigger, choke or voice steal. Its
  envelope now picks up from wherever it is, a choked hi-hat fades
  instead of cutting, and a note that takes over a sounding voice fades
  the old one out.

• The Limiter no longer clips transients. It now holds the gain
  reduction a peak needs across the whole lookahead and ramps into it, so
  a drum hit or a single-sample spike over the ceiling is turned down
  smoothly instead of being clipped. Latency is unchanged.

• Machines use about 20% less CPU. The fy compiler computes repeated
  expressions once, addresses state fields directly and writes them as
  soon as they are final. Output is bit-identical.

• The Extract pack finds uv when Slab is opened from Finder. Its
  download said uv was missing unless Slab was started from a terminal; it
  now looks in Homebrew's, ~/.local/bin and ~/.cargo/bin too.

• Starting a pack download can no longer crash Slab. Clicking Download
  or Import sometimes quit the app as the import began.

• Stop lets tails ring. Stopping releases the notes and lets reverbs
  and delays ring out instead of cutting everything with a click; press
  play again and the tails carry on.

• No clicks from edits while playing. Mute, solo, track enable and
  insert bypass fade over a few milliseconds. Adding, removing,
  reordering or replacing instruments and inserts fades the old chain
  out and the new one in, and a new instrument picks up held notes.
  Bypassing a lookahead or oversampled effect keeps its latency, so the
  rest of the mix no longer jumps in time; any latency change crossfades.
  Audio clips fade at cut edges and around seeks and loop wraps.

• 5 ms less latency. Audio reaches the output one block sooner on
  every path.

• About 6% less CPU for the same audio: peak scans run on vectors and
  are shared between inserts, compensation delays copy in bulk, and
  machines no longer copy their control tables every block.

• A blown-up machine can no longer reach your speakers. A track whose
  signal turns NaN or infinite (an unstable filter, a division by zero)
  is silenced for that block and its machines reset, so the fault cannot
  spread into buses, sends or the master. The output stage never passes
  a non-finite sample, in any clip mode; the hard clip used to turn one
  into full-scale DC. Exports write silence there too.

• No more stuck notes from mute, solo or odd notes. A track that is
  unmuted, un-soloed or re-enabled starts clean and picks up the notes
  sounding at that moment. Notes past a trimmed clip end, zero-length
  notes and notes shorter than a sample no longer hang.

• Smooth faders. Dragging a track or master fader ramps per sample
  instead of stepping each block.

• Tempo-synced effects on the master follow the song. A synced
  phaser on the master used to restart every block (a buzz); auditioning
  while stopped now runs synced LFOs too.

• Audio devices recover. If the input stalls or is unplugged while a
  track is armed, playback keeps running with silent input instead of
  freezing. A device that fails to open is retried about once a second.

• Adding tracks no longer races playback. Slicing to a sampler,
  extracting notes, chords or drums, placing stems and placing bounces
  insert tracks with the device stopped, as adding and moving tracks
  already did.

• Parallel rendering follows the output device. When macOS switches
  the default output (headphones, AirPods), playback renders serially
  until the render workers have joined the new device's workgroup. After
  a bounce, freeze or export they re-request realtime scheduling. Arming
  a track during a freeze or export no longer restarts the device.
  If workers keep the audio thread waiting for two blocks in a row,
  playback renders serially until they have been re-admitted.

• Fewer first-use stalls. Clip stretchers and the duplex capture
  buffer are allocated and touched before audio uses them.

---

Apple Silicon, macOS 13 or later. Built from v0.0.14, which is its complete source (GPL-3.0-or-later, see COPYING.md).

Install:

curl -fsSL https://github.com/nooga/slab/releases/latest/download/install.sh | bash

Or with Homebrew: brew install --cask nooga/tap/slab

Or open the DMG and drag Slab to Applications. It is not notarized, so the first launch needs System Settings > Privacy & Security > Open Anyway.