A sleek audio editor and mastering suite for Windows, built with WPF on .NET 10.
Lyrics & Speech (experimental) — Tools > Lyrics & Speech… transcribes a song or selected passage privately on your PC. Fine-tuned vocal isolation and Whisper large-v3 produce editable timed lines with replay/loop, a vocals-only audio extraction to a new tab for Save As/Export, uncertain-line review, full-recording checks for omitted phrases, marked recovered words and alternate readings, language/spelling hints, GPU acceleration with CPU fallback, and TXT/LRC/SRT/JSON export. The engine and models are included in the Release application and installer; no separate setup, account, or first-use download is required. See the quality limits and workflow. Export corrections before closing the audio tab.
Formats & files — sample-accurate WAV (16/24-bit PCM and 32-bit IEEE float) and AIFF/AIFF-C import, with native classic AIFF 16/24/32-bit PCM output, explicit dithered or undithered 16-bit output and Open As conversion to 16/24/32-bit; uncompressed DSD64/DSD128 import from DSF and DSDIFF with filtered 44.1 kHz floating-point working conversion; MP3/FLAC/M4A/WMA import; direct CD-DA track discovery, selection, extraction and import from Windows optical drives; export to WAV, AIFF, MP3, AAC, WMA and FLAC (where Windows provides the encoder) with bitrate choice, sample-rate conversion and selection-only export; batch converter with LUFS/peak normalization and effect-chain processing; recent files, session restore, autosave with crash recovery. AIFF-C and DSD are import-only. Imported AIFF-family files must be saved to a different .aif or .aiff path so Deep Groove cannot overwrite ancillary metadata it does not preserve; a DSD source always opens as a new PCM working copy and is never overwritten.
Editing — multi-resolution waveform with zoom to sample level, vertical amplitude zoom (Ctrl+wheel), and a per-channel 0/−3/−6/−12/−24 dBFS ruler with aligned grid lines; drag selection, lengthened or shortened afterwards by dragging either edge or with Shift+click; cut/copy/paste/delete/trim; unlimited undo/redo with a configurable memory budget and an Edit History panel (Ctrl+Shift+H) that lists every step applied to the file, marks where the document stands and where it was last saved, jumps back or forward across many steps in one action, and permanently discards a step and everything after it; markers and named regions with ruler flags, navigation shortcuts, a manager panel and sidecar persistence; edit-point smoothing (de-click joins); silence detection, trimming and split-to-regions.
Master rack — reorderable real-time effect chain with Gain & Trim, Level Normalizer, phase-safe Mono-to-Stereo and Stereo Width, Channel Balance & Alignment, Noise/Hiss Reduction, Hum Removal, Studio EQ, standards-based FM 50/75 µs and CD/IEC 50/15 µs emphasis/de-emphasis, Compressor, Noise Gate, Reverb, Stereo Delay, Chorus, Saturation, Low/High-Pass Filters and a 5 ms lookahead Precision Limiter. Every effect has bypass, remove, move and reset-to-default, with a global rack bypass that preserves individual settings; chains save/load as presets, including Vinyl Cleanup, Mono Record Presence, Clean Transfer, and three five-band Record-to-CD tone presets for gentle clarity, dull-source rescue, or warmth plus air. The transfer tone presets omit automatic denoising, reserve EQ headroom with trim, and finish with a −1 dBTP safety ceiling. Analyze & Tune examines representative quiet and musical passages, explains its confidence-backed recommendations, provides selective dry/tuned A/B preview, and builds custom Vinyl Cleanup or Clean Transfer settings without modifying the source or overwriting factory presets. The chain applies to a selection or renders to a new tab with latency compensation. True mono sources are promoted to stereo only when the enabled mono-to-stereo processor needs it. The enhancer also offers Natural Space, using short, diffuse reflections for subtle ambience on mono or existing stereo recordings while retaining the original mono sum and stereo detail. Start with Record to CD - Natural Space (28% amount, protected bass, reserved headroom and a -1 dBTP limiter); this creates space rather than reconstructing missing instrument positions.
Processing — gain, normalize, equal-power fades, reverse, DC removal, silence insertion; Match Loudness across open tabs, levelling a set to a delivery preset (EBU R128, ATSC A/85, Apple Music, streaming, CD master) or to the selection's own quietest, average or reference track — measured once with BS.1770 gating and true peak, re-planned instantly when the mode changes, applied as gain alone with the true-peak ceiling honoured and any shortfall reported as the limiting it would have taken, one undoable entry per file naming the level it matched to; channel tools (swap, phase invert, balance, mono↔stereo, channel extraction); WSOLA time stretch and pitch shift; windowed-sinc sample-rate conversion.
Making a CD from finished files — Audio CD from Files… takes a set of separate masters, brings them onto the CD's clock once, lays them end to end as one programme with one track each and opens the CD window with two seconds between every pair, ready to reorder, retitle, retime or regap before Export writes sector-aligned 16-bit WAVs and a CUE sheet with real pregaps (INDEX 00/INDEX 01). Export the burner package either way: WAV + CUE writes one file per track, indexed from each file's own start; Image + CUE writes the whole programme as a single WAV with a cue sheet whose INDEX times are positions on the disc and climb down the sheet - the form a ripper produces, and the one a burner that reads only a single FILE line needs. Both describe the same disc, sample for sample. The gap here only ever adds: a finished master keeps its own opening and ending, where the vinyl transfer path deliberately trims the record's own uneven quiet back to the music first. Needs no open file and no montage.
Restoration & CD transfer — a source-aware vinyl transfer workbench distinguishes an ordinary RIAA-equalised recording from a flat cartridge capture. Flat mode first consumes raw-domain clipping and click plans while their rail heights and coordinates are still exact, then removes measured DC, corrects supported azimuth, repairs crackle, corrects validated wow, applies the selected playback curve exactly once, and finally performs rumble, vertical-noise, hum and hiss cleanup in the equalised domain. Impulses are scanned across the requested range; continuous cleanup uses a bounded representative excerpt centred on a quiet passage, repaired and prepared in the same order as the render. Wow analysis masks detected damage without cloning the side, and weak measurements remain report-only. Disc-curve state follows undo/redo, rendered copies and bit-depth/channel conversions, and is embedded in WAV, AIFF and Wave64 so a reopened file cannot accidentally receive RIAA twice. Gentle/Balanced/Strong presets, bypass A/B and a bounded audition leave the source untouched until one undoable apply; wow is explicitly excluded from the cropped audition because only a whole-file render has its cumulative time position. Output defaults to 100% restored, flat mode prevents an invalid unequalised/equalised blend, and its CD shortcut is withheld until the separate loudness/true-peak decision is complete.
Analysis — real-time spectrum analyzer, offline spectrogram, loudness history graph, goniometer with stereo correlation and balance; peak/RMS meters with hold; EBU R128 momentary/short-term/integrated LUFS, LRA and true peak; per-file audio statistics with clipping count; pitch detection (tuner) and tempo/BPM estimation.
Recording — WASAPI capture with device picker and live peak/RMS/hold meters; the KORG DS-DAC-10R is recognized by both its product and Windows driver names, with an in-context reminder and one-click launcher to set hardware gain and confirm the current Line/Phono mode in KORG's Setting Tool (AudioGate changes that mode), plus a truthful handoff for native ASIO DSD capture (record DSF/DSDIFF in AudioGate, then edit it here). A vinyl-oriented Recording Level Assistant runs a discard-safe preflight scan, measuring intersample true peak, a click-resistant programme peak (isolated pops—even rail-touching ones—can't set the gain for a whole side), projected peak with a crest-aware unseen-transient reserve, program RMS and integrated LUFS, noise floor, channel balance, DC offset, 50/60 Hz hum and sustained clipping — then recommends a linked input-gain reduction or an optional conservative increase before the take, and applies it in one click to the Windows device level and Deep Groove fine trim, reporting any residual a coarse driver step leaves behind. The safety reserve is evidence-based — it never falls below a schedule set by scan length, and rises above it when the loudest moment sits well above the peak the recommendation is built on or when the running maximum is still climbing late in the scan. Reserve limits increases without forcing programme already at or below the target ceiling lower, and an explicit whole-side mode samples from lead-in through run-out while holding the safest result. The target ceiling is yours to set — anywhere from −1 to −12 dBTP in half-decibel steps, default −6 — because declicking and peak reconstruction can put a repaired peak back above what was captured; it lives both in Settings and on the assistant's own header, where changing it re-derives a running scan from what it has already measured rather than restarting it. A scrolling 30-second input-history strip shows the scan against the target band; the safest outcome is remembered per input device together with the setting that realizes it, so a later session can restore or forget it; and the check summary travels into the take's metadata. Isolated stylus clicks never dictate the gain for a whole side, but clipping dense enough to define a block is reported as overload regardless. The monitor stream switches atomically into retained recording so calibration audio never enters the take. An automatic needle-drop mode listens while armed, starts on the stylus-contact click, and preserves a 250 ms safety pre-roll so the opening groove is intact. Run-out auto-stop learns the side's own groove floor, defers its wait while a slow fade is still falling, and ends the saved take four seconds after the confirmed ending. A transport Arm control starts the selected persisted input immediately when Record is pressed; metronome count-in at any BPM; click track while recording; punch-insert into the current selection with smoothed joins. Capture finalization waits for the device tail, keeps interrupted takes recoverable, and marks every new recording as unsaved.
Workflow — advanced audio-hardware settings with explicit or Windows-role-following input/output routing, WASAPI shared/exclusive modes, event/poll scheduling, independent 3–500 ms playback/capture buffer requests, live endpoint format/period/volume/exclusive-rate diagnostics, refresh, a non-destructive output-path tone test and a discard-only input peak/RMS test; autosave and export defaults with restore-defaults; searchable command palette (Ctrl+Shift+P); window placement memory; CPU/RAM readouts; drag-and-drop and command-line file open.
Finding a stubborn ringing click — roughly select up to 10 seconds around the sound, then use Find defect beside Heal (also in the Restore menu). It highlights a possible time/frequency repair area and zooms to it in the split view. Listen, then press Heal to apply an undoable repair. Find defect only selects; it does not change the recording.
Requires the .NET 10 SDK (ships with Visual Studio 2026).
dotnet build WaveLab.sln -c Release
dotnet run --project src/WaveLab
Space play/pause · Home go to start · Ctrl+O/S/Shift+S open/save/save-as · Ctrl+E export · Ctrl+Z/Y undo/redo · Ctrl+Shift+H edit history · Ctrl+X/C/V/Del edit · Ctrl+A select all · Ctrl+= / Ctrl+- / Ctrl+0 zoom · Ctrl+M marker · Ctrl+Shift+M region · Alt+←/→ marker navigation · Ctrl+R record · Ctrl+Shift+P command palette · Shift+click extend selection · wheel zoom · Shift+wheel pan · Ctrl+wheel amplitude zoom
UI typefaces: Inter and JetBrains Mono, embedded. Design mockups and the feature roadmap live in docs/.