master_core imported safe_title from font_manager, which imports PyQt5 at
module load. The web extra installs no Qt, so `ujm-web` crashed on startup
with "No module named 'PyQt5'" inside the container. safe_title is only a
UTF-8 round-trip, so master_core now carries its own Qt-free copy; the
desktop path keeps using font_manager's.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_015UXA39Sh9coAx4FhRkd8R2
Multi-stage image built from the `web` extra only, so it carries no Qt: the
whole web path (webapp, web_session, metrics_json, and the shared analysis
modules) imports no Qt, pyqtgraph or matplotlib canvas. The build stage has the
compiler scipy/numba need; the runtime layer installs the built wheel, so the
same image works from a checkout or a release tarball.
Compose declares no audio volume on purpose. The app keeps no audio, so there is
nothing to persist: `read_only` plus a /tmp tmpfs means request spools are
memory-backed and never reach a disk. Publishing is loopback-only because the
tool has no authentication, and BLAS thread pools are pinned because analysis is
CPU-bound.
Fix three packaging bugs the build surfaced, each of which broke the image:
- `pip install --extra web` is not valid for a source tree; uv wants `.[web]`
and failed with "Requesting extras requires a pylock.toml".
- The wheel shipped no static assets at all. `package-data` keys match
*packages*, and the app modules are top-level `py-modules`, so the `"*"` key
matched nothing - verified by inspecting the built wheel: 14 .py files, zero
HTML/JS/CSS. An installed `ujm-web` would have 404'd every asset. `static/` is
now declared a package (hence its packaging-shim `__init__.py`) so the assets
install next to the modules, which is where `webapp.py` looks for them.
- The Dockerfile's curated COPY list omitted modules that `py-modules` requires,
failing the build with "file <module>.py not found". It copies the checkout,
filtered by .dockerignore.
Move PyQt5/pyqtgraph into a `desktop` extra so the web install is genuinely
Qt-free. Verified: `find` for Qt in a web-only venv now returns nothing, and the
desktop path still resolves and imports.
Add scripts/make_fixtures.py so the smoke suite has no network or ffmpeg
dependency (the tagged MP3 is optional and its assertions skip when absent), and
make the suite tolerate that. Verified against the *installed package* running
the container's own command line, from a directory with no source tree:
61/61 passing.
Also correct README claims that had gone stale: BPM (removed), the matplotlib
canvas (replaced by pyqtgraph), and font_control_widget.py (deleted).
Reuses the analysis backend unchanged. metrics.py already split compute()
from build_spec(), and plotspec.PlotSpec is backend-agnostic, so the web UI
reimplements only the renderer: uPlot for curves/bands, a canvas for the
spectrogram heatmap. No PyQt5, pyqtgraph or matplotlib on the web path.
Uploaded audio is never persisted on the host. The browser copies each file
into Origin Private File System (sandboxed browser storage) and streams the
bytes to the server per request; the server spools each body to a temp file,
decodes, computes, and unlinks it in a finally before responding. There is no
upload directory, no cache of audio, and no session state keyed on a path:
file_id is the SHA-256 of the audio, so re-sending a master for another metric
reuses the first decode. Verified: a spool is observable mid-request and gone
after the response.
Being precise: the bytes do have to reach the server, since librosa decodes
from a path and the analysis is CPU-bound numpy/scipy. The claim is that
running this does not accumulate masters on the host, not that audio is never
on the machine mid-request.
New:
metrics_json.py PlotSpec/metric data -> JSON (the renderer seam)
web_session.py headless analysis session, content-addressed tracks
webapp.py FastAPI routes, raw-body analysis, ujm-web entry point
static/ browser front end + test suite
scripts/webapi_smoke.py 61 end-to-end assertions against a running server
Also fixes a bug the direct-uvicorn dev loop had hidden: the CLI advertises
logger_setup's WARN level but passed it straight to uvicorn, which only knows
'warning' -- `ujm-web --log-level WARN` died with KeyError: 'warn'.
Removes an earlier /api/collect debug route that served every source module
over HTTP.
pyproject: 'web' extra (fastapi, uvicorn, python-multipart), package-data for
static/, Apache-2.0 licence.
- Replace LICENSE with properly filled Apache-2.0 (Mikkeli Matlock 2026)
- Add license field to pyproject.toml
- Add SPDX-License-Identifier headers to all .py source files
LUFS integrated + LRA:
- Reimplement BS.1770 two-stage gating (integrated) and EBU 3342 (LRA) directly
from the cached K-weighted signal, dropping pyloudnorm's two whole-signal
re-filters. Validated bit-equal to pyloudnorm across steady/dynamic/quiet
signals (0.0000 diff). LUFS compute ~1.9s -> ~0.6s (orig 3.8s). pyloudnorm now
only supplies the filter coefficients.
Prefetch on load:
- After a file loads, PrefetchWorker computes the remaining metrics in the
background (sequential, cooperatively cancellable, skips on-demand hits), so
the first switch to any metric is instant. Superseded when a new file loads.
Status slip:
- Workers measure compute time; metricTiming + phase/duration progress messages
drive the slip ("Loaded in Ns - computing X...", "X computed in Ys"). The old
fake percentage (10% then done) is logged but no longer shown.
- shutdown() stops background threads on window close.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Profiled hot spots on a 4-min track and cut the worst offenders:
- Remove BPM: librosa.beat.beat_track ran on every load (~3.7s) for a number no
better than tapping by hand. Dropped from AudioFile + the metadata panel.
- LUFS short-term: replace 474 per-window pyloudnorm.integrated_loudness calls
with one K-weighting pass (reusing pyloudnorm's own filter coefficients) + a
vectorised sliding mean-square. This is true *ungated* EBU R128 short-term
(the old loop wrongly gated each 3s window). Integrated + LRA still use
pyloudnorm's gated calls. ~3.8s -> ~1.9s.
- PSR: reuse LUFS's short-term series (memoised on the AudioFile) + vectorised
sample-peak. ~3.0s -> ~0.2s.
- True Peak: oversample the whole signal once, then an O(N) running max over
windows instead of per-window resample_poly. Bit-identical to the old loop
(max|diff| 0.0000 dB). ~2.1s -> ~1.1s.
- Crest Factor: peaks via the same O(N) running max (last per-window loop gone).
lufs+psr+true_peak: ~9.2s -> ~3.2s, plus ~3.7s of BPM removed from every load.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Reference lines:
- Kept per metric (dict keyed by metric_id) so switching metrics and coming back
preserves them, instead of clearing on every switch
- Spectrogram lines read/edit/drag in Hz: the renderer installs Hz<->row-index
transforms (the heatmap y-axis is a row index), so value, label, the edit
dialog, and the new-line default all speak frequency. Curve metrics stay
identity. Round-trip verified (1000 Hz -> row -> 1000 Hz)
- Line label and drag-readback always in the metric's natural units
Controls:
- Relative-time abs/rel is now a checkbox toggle, not a dropdown
- Removed the font control panel; UI font is locked to M PLUS 1 Code @ 10pt via
font_manager.apply_fixed_font (system-default fallback). Deleted
font_control_widget.py
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Reference lines:
- Side-panel cluster (ref_line_widget.py): list of lines with Add/Edit/Remove/
Clear and a properties dialog (value, colour, line style, tag)
- Lines own their state as RefLineProps (held by MainWindow), drawn with a
triangle drag-handle; dragging writes the value back and refreshes the list
- Replaces the old centre-only add-and-drag-the-whole-line approach
Comparison:
- Time-axis mode selector: Absolute (seconds) vs Relative (% of each track's own
length), so a long track and a short one line up by song position. Pure view
transform (plotspec.apply_x_mode), no recompute
- Overlaid different-length tracks now autorange their x union in absolute mode
instead of clipping to the first track's span
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
- SpectrogramMetric: log-frequency STFT power heatmap over time, magma
colormap, -80 dB floor. Adaptive hop caps time bins at ~4000 so long
tracks stay responsive on redraw; N_FFT=4096 keeps low-freq resolution.
- master_core: load audio at native sample rate (librosa.load sr=None)
instead of librosa's 22050 Hz default, so the full band up to the
file's own nyquist (~22 kHz at 44.1 kHz) is analysed. ~2x heavier on
44.1/48 kHz files, by design.
- metrics: shared _show_axis_extents helper forces each axis's exact
min/max onto the tick list with compact labels (_fmt_tick), so the
true range is always readable -- notably the spectrogram's 22 kHz top,
which otherwise sits unlabelled between log-scale decade ticks. Applied
to all metric renders.
- Docs: README + CLAUDE updated for the new metric, native-rate loading,
and axis-readability behaviour.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Three new metrics plug into the existing Metric registry. All inherit
the async compute path, so first-use is non-blocking and subsequent
switches hit the per-(file, metric) cache.
- CrestFactorMetric: 20*log10(peak/RMS) per 1 s window, 0.25 s hop.
Uses cumsum-of-squares for O(N) RMS. Reference lines at 12 dB
(roomy) and 6 dB (heavily limited).
- PSRMetric: sample-peak (dBFS) minus short-term LUFS over the same
3 s window as LUFSMetric, so the two plots line up. Reference lines
at 10 LU (good punch) and 4 LU (squashed).
- TruePeakMetric: ITU-R BS.1770 oversampled true peak via
scipy.signal.resample_poly at 4x over 250 ms / 100 ms windows.
data["integrated_tp_db"] carries the track-wide max.
- LUFSMetric now also computes Meter.loudness_range and surfaces it
in the plot legend alongside the integrated value.
- Shared _to_dbfs helper floors 20*log10 at _EPS so silent windows
don't explode.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Introduce a Metric ABC (compute on worker thread, render on GUI thread)
with a METRICS registry, and refactor the analysis pipeline around it.
RMS power (ported), raw waveform, and BS.1770 LUFS (via pyloudnorm) ship
as the initial three; new metrics drop in by appending to METRICS.
- metrics.py: Metric ABC + RMSPowerMetric, WaveformMetric (locked to
+/-1.1 y-range for float headroom), LUFSMetric (short-term 3 s window
+ integrated value, with streaming-target reference line).
- plot_control_widget.py: metric selector dropdown + Refresh Plot button
(moved out of FontControlWidget).
- analysis_results_manager.py: AnalysisResult caches the AudioFile and a
per-metric data dict; new MetricComputeWorker runs metric switches off
the GUI thread via metricComputeStarted/metricReady/metricComputeError
signals, so LUFS on a 12-minute track no longer stalls the UI.
- main.py: all redraw paths funnel through one _render_or_request helper;
stale-result guards keep slow computes from overwriting fresh selections.
- plotting_engine.py removed (metadata text moved onto AnalysisResult;
figure construction lives in each Metric).
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
- Reflect current implementation status with complete GUI and font system
- Update roadmap with prioritized development plan focusing on plot control widgets
- Reorganize future goals into urgent/short-term/mid-long term categories
- Add comprehensive feature overview and usage instructions
- Clarify next priority: plot control system clustering
🤖 Generated with [Claude Code](https://claude.ai/code)
Co-Authored-By: Claude <noreply@anthropic.com>
- Add open file button for direct audio file selection
- Implement clustered FontControlWidget with font selector, size slider, and refresh controls
- Add auto-regeneration of matplotlib plots when fonts change
- Enhance FontManager with system font discovery and startup selection logic
- Replace individual font selector with comprehensive font control interface
- Support both custom fonts and system font candidates with proper prioritization
🤖 Generated with [Claude Code](https://claude.ai/code)
Co-Authored-By: Claude <noreply@anthropic.com>
Major refactor from popup-based to persistent PyQt5 interface:
- Extract plotting logic from AudioFile class into separate PlottingEngine
- Create AudioVisualizationWidget with embedded matplotlib canvas
- Add AnalysisResultsManager as bridge between processing and GUI
- Replace simple drag-drop widget with professional splitter layout
- Preserve legacy batch processing mode with execution guard
Features:
- Drag-and-drop audio analysis (.mp3/.wav/.flac support)
- File list with metadata display (BPM, amplitudes, track info)
- Persistent visualization area (no more matplotlib popups)
- Multi-file support with click-to-view functionality
- Threading-ready architecture for future background processing
Technical improvements:
- Clean separation of concerns (analysis/visualization/GUI)
- Qt signal-slot communication pattern
- Modular component design ready for multithreading
- Proper import guards prevent legacy code interference
🤖 Generated with [Claude Code](https://claude.ai/code)
Co-Authored-By: Claude <noreply@anthropic.com>
- Added CLAUDE.md with detailed project documentation and roadmap
- Enhanced README.md with usage section for command line and GUI modes
- Updated with Claude Code credit
- Improved master_core.py with better song name handling and BPM display
🤖 Generated with [Claude Code](https://claude.ai/code)
Co-Authored-By: Claude <noreply@anthropic.com>
mikkeli
irene
Mikkeli Matlock