- Remove serial.SerialException from except clause since 'serial'
is not defined when import serial fails, causing UnboundLocalError
- Catch only ImportError and OSError
- GET /api/audio/profiles — list available profiles with params
- GET /api/audio/profile — current profile + jack_running + xrun_count
- POST /api/audio/profile {profile: 'standard'|'low'} — switch profile,
restart JACK server + JACK audio client atomically
- Adds audio_system + jack_audio refs to WebServerDeps
- Rollback on JACK restart failure
Complete rewrite of _build_nam_model and _build_wavenet to handle:
- SlimmableContainer (A2): pick submodel by quality, recurse into WaveNet
- WaveNet: proper dilated convs with causal padding, 1x1 layer1x1 post-conv
- Linear: simple pass-through with gain/bias
Weight import matches NAM export order:
rechannel -> per-layer(conv w+b, input_mixer w, layer1x1 w+b) -> head_rechannel w+b -> optional head -> head_scale
Also fixes MOCK_PRESETS in frontend to match actual factory preset names.
The PyPI 'nam' package (v0.0.3) is Neural Additive Models for tabular
data — completely different from Neural Audio Model (guitar amp modeling).
Implemented a local _init_from_nam() that builds a WaveNet NAM model
from the Core ML .nam JSON format. Uninstalling the wrong package.
Reordered UI_DIST_DIR candidates so pi-multifx-pedal-ui/dist (the
correct Helix-style pedal UI) is checked first. The old priority
put src/web/ui-dist first, which allowed wrong-build deploys to
mask the real UI.
Also add UI_DIST_DIR env var to systemd service for explicit path.
Two issues found and fixed in the deployed UI bundle:
1. String.fromCharCode(64+y) where y is 0-based bank index
→ 64+y gives 64+0=64 = '@' for bank 0
→ Fixed: 65+y gives 65+0=65 = 'A' for bank 0
2. e.bank||1 where falsy 0 falls through to 1
→ Shows 'Bank 1' for bank 0, 'Bank 2' for bank 1
→ Fixed: e.bank||0 so 0 renders as 'Bank 0'
Source code (frontend-react/src/App.jsx) was already clean —
the bug was in a transport/song mode feature that was built
from uncommitted changes then reverted.
Adds snapshot system to Helix Stadium:
Backend:
- Snapshot data model (8 slots per preset, captures block bypass
+ params + master volume)
- Snapshot API endpoints: list, save, recall, rename, delete
- Integrated into preset serialization/deserialization
- Tracks current_snapshot in state / WebSocket broadcasts
Frontend:
- Camera icon in header bar with current snapshot number
- SnapshotPanel as slide-up overlay
- 8 snapshot slots in a horizontal row
- Tap to recall filled snapshot, tap empty slot to save
- Press-and-hold (600ms) to save current state
- Inline name editing on tap
- Discard Edits toggle
- Green flash indicator on save
- Delete button for active snapshot
- New Preset button: finds first empty slot, saves with default name
- Looper/Recorder tab: Record/Stop via arecord backend, list captures, play back
- Rig: shows loaded IR name more clearly + quick-link buttons to browse IRs/Models
- Backend: capture/start, capture/stop, captures API endpoints with file serve
- Status bar: CPU load + sample rate from /api/state, TUNER badge shown immediately
- Rig: optimistic toggle updates (bypass/tuner respond instantly), prominent tuner card
- FX Chain: loads real block params from /api/block-params/{fx_type}, Add Block picker with all FX types
- Record: replaced fake multitrack with real Signal Monitor (live I/O VU, clip detect, level history)
- Backend: added _gather_system_stats() returning cpu_percent and sample_rate
- Added input_level and output_level to /api/state (scaled 0-100)
- React RigScreen now reads real levels from state instead of random walk
- When no audio flows, VU meters show flat (0) instead of fake animation
- New dark UI with 7 tabs: Rig, FX Chain, Models, IRs, Presets, Settings, Record
- All screens wired to live API endpoints
- Real-time state via WebSocket + polling fallback
- Models screen: list/load/unload/upload NAM files + Tone3000 search/install
- IRs screen: list/load/unload/upload IR files + Tone3000 search/install
- Presets screen: browse banks, activate, save, delete
- Settings screen: WiFi scan/connect/hotspot, Bluetooth/MIDI, 4CM routing
- Rig screen: live state, volume, bypass, tuner, signal chain
- FX Chain screen: block params, per-block bypass
- Serves at / instead of /ui
- Built with Vite, ~71KB gzipped
Add UI_DIST_DIR auto-detection and /ui static mount in the
FastAPI server so the built React frontend is served alongside
the main dashboard.
- UI_DIST_DIR env var override
- Auto-detect common project layouts (frontend-react/dist,
pi-multifx-pedal-ui/dist)
- Mount with html=True for SPA client-side routing
- Add JackAudioClient class to src/system/audio.py — real-time JACK
client owning ports and a process callback that streams audio through
the DSP pipeline. Handles mono (1ch I/O) and stereo 4CM (2ch I/O).
- Wire PedalApp.boot(): read audio.mode from config, set pipeline
routing_mode (mono/4cm), create and start JackAudioClient with
matching channel count.
- Add PedalApp.set_routing_mode(mode) — hot-switch between mono and
stereo_4cm at runtime: updates pipeline.routing_mode, restarts JACK
client with new channel count via set_channel_count(), re-connects
JACK ports.
- Wire shutdown: stop JackAudioClient before MIDI/footswitch.
- Export JackAudioClient from src.system.__init__.
- 53 tests pass (22 integration + 31 audio).
Pipeline:
- process() dispatches to _process_mono() or _process_4cm() based on routing_mode
- _process_4cm() splits chain at routing_breakpoint: pre blocks on ch0, post on ch1
- _process_single_block() extracted for reuse in both mono and 4cm paths
- routing_mode/routing_breakpoint load from preset via load_preset()
- set_routing() for runtime configuration
- Properties: routing_mode (mono|4cm), routing_breakpoint with validation
Web server:
- GET /api/routing — current routing mode and breakpoint
- POST /api/routing — set routing mode/breakpoint, persist to current preset, WS broadcast
- _gather_state() includes routing_mode and routing_breakpoint
Web UI:
- Settings page: 4CM toggle + breakpoint slider with routing description
- Dashboard: routing badge (4CM/Mono) with breakpoint info
- app.js: WebSocket handler, updateRoutingUI(), toggle4cm(), set4cmBreakpoint()
- style.css: .badge, .badge-4cm, .badge-mono, .routing-status, .routing-desc
Tests: mock pipeline fixture updated with routing_mode/routing_breakpoint
- Add routing_mode (mono/4cm) and routing_breakpoint (default 7) to Preset
- Serialize/deserialize in _preset_to_dict / _preset_from_dict
- Backward compatible — old presets default to mono/7
- Add get_4cm_config() / set_4cm_config() helpers on PresetManager
- set_4cm_config validates mode and saves back to disk
- All 48 related tests pass (46 preset + 2 integration activation)
- Full FFT overlap-add IR convolution in IRLoader (process(), set_mix(), toggle)
- Lazy FFT computation — IR FFT padded to correct block+ir size on first process()
- Wet/dry mix control, enabled/disabled toggle with tail clearing
- Fixed pipeline._apply_ir_cab() to delegate to IRLoader.process() instead of
poking internals (old code had array-size mismatch bug: IR FFT at ir_len vs
block FFT at conv_size)
- 46 tests: loading, convolution correctness, overlap-add state, mix, toggle,
directory listing, performance budget (all <5ms even at 8192 taps), edge cases
- scripts/download_irs.sh: free IR pack downloader (God's Cab, Seacow)
- 15 FX blocks implemented with per-block state isolation
- All blocks <500us per 256-sample block (reverb closest at 465us on x86)
- 57 unit tests all passing (per-effect, chain, bypass, state isolation)
- Benchmark script at scripts/benchmark_fx.py
- Vectorised delay reads via read_block_varying()
- scipy.lfilter for EQ (3-band RBJ) and reverb damping
- Fixed _DelayLine.write_block wraparound crash for large blocks
- Comb/Allpass buffers sized to BLOCK_SIZE * 2 minimum
shawn