v10.2.0

What's New

preservesPitch prop on MediaElementPlaylistProvider

The HTMLAudioElement.preservesPitch was previously hardcoded to true. It's now configurable via a new boolean prop (default: true, backwards compatible).

<MediaElementPlaylistProvider
  track={trackConfig}
  preservesPitch={false}  // disable browser pitch correction
/>

Set to false when using an external pitch processor (e.g., SoundTouch AudioWorklet) that handles pitch compensation itself.

MediaElement example update

The Media Element example now includes a Preserve Pitch toggle so you can hear the difference between pitch-preserved and non-pitch-preserved playback rate changes.

Full Changelog: v10.1.2...v10.2.0

v10.1.2

What's Changed

Improvements

• Migrate browser package from Vite to tsup — Consistent build tooling across all 13 packages. tsup auto-externalizes dependencies and peerDependencies, preventing the class of bundle drift that caused #317.
• Remove manual external lists from all tsup configs — Redundant with tsup's auto-externalization. Removes 52 transitive devDependencies (Vite, Rollup, etc.). ESM bundle reduced from 398KB to 251KB.

Full Changelog

v10.1.1...v10.1.2

v10.1.1

What's Changed

Bug Fixes

• Fix clip drag not constrained horizontally — The Vite external list referenced stale @dnd-kit v4 packages (@dnd-kit/core, @dnd-kit/modifiers) instead of the v0.3.x packages actually used (@dnd-kit/abstract, @dnd-kit/dom, @dnd-kit/react). This caused @dnd-kit to be bundled into the npm output, creating duplicate instances at runtime that broke RestrictToHorizontalAxis. (#317)

Improvements

• Derive Vite externals from peerDependencies in package.json automatically, preventing this class of drift in the future.

Full Changelog

v10.1.0...v10.1.1

v10.1.0

What's Changed

Features

• Web Audio routing for MediaElementTrack — MediaElementPlaylistProvider now accepts an optional audioContext prop to enable Web Audio routing through a gain chain (HTMLAudioElement → MediaElementSourceNode → fadeGain → volumeGain → destination). This enables:
  • Fade in/out with 4 curve types: linear, logarithmic, exponential, sCurve
  • Tone.js effect bridging via native→Tone.Gain connection pattern
  • Partial fade curve preservation when seeking mid-fade

Improvements

• Replace hand-rolled AudioWorklet types with @types/audioworklet
• Consolidate FadeConfig type alias in @waveform-playlist/core (re-exported by playout and media-element-playout)
• Fix stale API references across documentation for v10

Full Changelog

v10.0.0...v10.1.0

v10.0.0

v10.0.0

Breaking Changes

@waveform-playlist/recording:

• Removed UI component exports: RecordButton, MicrophoneSelector, RecordingIndicator, VUMeter (and their prop types). Use SegmentedVUMeter from @waveform-playlist/ui-components and build your own recording controls.
• useMicrophoneLevel options changed: removed fftSize and smoothingTimeConstant, added channelCount. Now uses AudioWorklet-based metering instead of Tone.js Meter.
• Recording worklet files moved from @waveform-playlist/recording to @waveform-playlist/worklets (auto-installed as a dependency).

New Package

• @waveform-playlist/worklets — Shared AudioWorklet processors for sample-accurate metering (meter-processor) and multi-channel recording (recording-processor).

New Features

• SegmentedVUMeter — LED-style meter component with horizontal/vertical orientation, configurable segments, dB range, color stops, per-channel peak hold, and dimmed inactive scale
• Worklet-based metering — Sample-accurate peak and RMS measurement replacing Tone.js Meter polling
• useOutputMeter — Master bus output metering hook with auto-reset on stop
• Multi-channel useMicrophoneLevel — Auto-detects mono/stereo from mic, mirrors mono to fill requested channel count, exposes per-channel levels, peakLevels, rmsLevels
• Overdub recording — Record over existing audio with latency compensation (outputLatency + lookAhead). Timeline position captured at record-start.
• Incremental track addition — engine.addTrack() avoids full playout rebuild when appending tracks
• DAW-style toolbar — Transport strip + meter bridge with IN/OUT meters, mic selector, master volume, export
• Recording keyboard shortcuts — R to record, Shift+R to create new track and record
• indefinitePlayback — Timeline fills visible scroll container
• Device hot-plug detection — Auto-detects plugged/unplugged microphones, falls back to first available device

Error Handling

• Meter setup failures exposed via error state
• Short recordings set user-facing error instead of silent discard
• Consistent console.warn('[waveform-playlist] ...') logging
• All recording hooks use getGlobalContext() from playout (CLAUDE.md pattern 22)

Test Coverage

• 46 new unit tests: incremental track addition, latency compensation, mono-to-stereo mirroring, SegmentedVUMeter activation

v9.5.2

Bug Fixes

• Fix duplicate audio on play/pause/play — Add event.repeat guard in keyboard shortcuts to prevent held-key repeats from firing multiple concurrent play() calls during async engine initialization
• Fix engine rebuild race condition — Add deferEngineRebuild={loading} to Flexible API example to prevent per-track engine rebuilds during progressive loading

Tests

• Add 16 unit tests for keyboard shortcut handling including bug reproduction scenario

Documentation

• Document jsdom keyboard event testing quirks in browser package CLAUDE.md
• Document deferEngineRebuild requirement for multi-track examples in website CLAUDE.md

v9.5.1

Spectrogram Performance & Reliability

Performance

• Worker pool — Parallel per-channel FFT computation (~1.5s instead of ~2.9s for stereo). Pool size defaults to 2; configurable via workerPoolSize prop. FFT fan-out capped to actual channel count (excess workers sit idle).
• Three-tier rendering — Viewport-first paint, then buffer zone, then background batches
• Generation-based abort — Cancels stale FFT requests on scroll, preventing queue blocking
• LRU cache — 16-entry FFT cache prevents recomputation on scroll-back
• Lazy per-batch FFT — Bounded memory per render batch, prevents OOM on 1hr+ files
• SpectrogramAbortError — Type-safe abort detection via instanceof instead of string matching

Cleanup

• Fix canvasIds[i] undefined variable in DPR scaling error path
• Remove dead spectrogramDataMap from SpectrogramIntegration interface
• Use string interpolation for all console.error calls

Documentation

• Update llm-reference.md with current SpectrogramWorkerApi interface
• Add workerPoolSize, SpectrogramAbortError, createSpectrogramWorkerPool to docs
• Add spectrogram architecture decisions to packages/spectrogram/CLAUDE.md

v9.5.0

What's New

• Multi-channel recording pipeline — Recording now captures all channels from the microphone stream with per-channel peaks. Channel count is auto-detected from the stream via getSettings().channelCount.
• Configurable peak bit depth — RecordingOptions accepts bits?: 8 | 16, flowing through recordingState.bits to the renderer.

Bug Fixes

• AudioWorklet buffer overflow — At 44100Hz, the 128-sample AudioWorklet quantum doesn't divide evenly into the ~16ms buffer (705 samples), causing ~9% of recorded samples to be silently dropped. Fixed with a loop that handles frame boundary crossings.
• Real AudioContext sample rate — useAudioTracks now passes the actual audioContext.sampleRate to track building instead of always falling back to a hardcoded 48000.
• Canvas flicker — Channel, PianoRollChannel, and SpectrogramChannel switched from useEffect to useLayoutEffect for canvas drawing, preventing clearRect from being visible for one frame.
• Recording preview width sync — durationSamples uses duration * sampleRate (zoom-independent) instead of peaks-derived calculations that depend on samplesPerPixel.
• Mono consistency — Live preview and post-recording paths both respect the mono flag consistently.

Full Changelog: v9.4.1...v9.5.0

v9.4.1

Bug Fix

• Fix audio bleeding between pages — TonePlayout.dispose() now stops the Tone.js Transport and all active AudioBufferSourceNodes before cleanup. Previously, audio continued playing through the global AudioContext after provider unmount (e.g., during SPA client-side navigation). (#312)
• Resilient cleanup — stop() and dispose() now guard per-track operations with individual try-catch blocks, ensuring one track failure doesn't skip cleanup of remaining tracks.

v9.4.0

What's New

• renderPlayhead prop for MediaElement — MediaElementWaveform and MediaElementPlaylist now accept a renderPlayhead prop, matching the existing API on Waveform. Pass PlayheadWithMarker or a custom component for a triangle-marker playhead.

• Hook isolation for custom playheads — Both PlaylistVisualization (WebAudio) and MediaElementPlaylist now wrap renderPlayhead calls in dedicated components (CustomPlayhead, CustomMediaElementPlayhead), preventing React's "Rendered more hooks" error when the prop is conditionally provided.

• Time display fix — Media element example now uses currentTimeRef with a requestAnimationFrame loop for smooth 60fps time updates during playback.

• Stem tracks custom playhead — Stem tracks example now uses PlayheadWithMarker to exercise the WebAudio custom playhead path.

• Media Element Playout guide — New documentation page covering usage, custom playheads, and the 4 context hooks.

Full Changelog

v9.3.3...v9.4.0