Add experimental toggle to force PipeWire clock.force-quantum alongside
clock.force-rate for true bit-perfect playback. Quantum is matched to
sample rate (1024 for <=48kHz, 2048 for <=96kHz, etc). Both values
reset to 0 on stop and pause-suspend to avoid affecting other apps.
Requires DAC Passthrough to be enabled, PipeWire backend only.
Added DEPRECATED headers to src-tauri/src/player/mod.rs and
src-tauri/src/queue/mod.rs indicating these will be removed
after migration to qbz-player crate is complete.
New code should use:
- qbz_player::Player (via CoreBridge -> QbzCore)
- qbz_player::QueueManager (via CoreBridge -> QbzCore)
Add a toggleable gapless button next to the repeat button in the
NowPlayingBar. The setting persists in audio_settings.db and controls
whether the audio thread pre-queues the next track for gapless
transitions. Includes i18n keys for all 4 locales.
Backend: Add PlayNext command to audio thread that appends decoded
sources to the existing Rodio Sink for seamless transitions. The audio
thread detects when ~5s remain and signals gapless_ready via SharedState.
Position-based transition detection updates track metadata when the
queued source begins playing.
Frontend: playerStore detects gapless_ready flag, invokes
play_next_gapless (cache-only lookup), and handles transitions by
advancing the queue without stop/play cycle. playbackService gains
gaplessTransition option to skip stop_playback when backend already
transitioned.
Gapless is only eligible when: same sample rate/channels, cached data,
Rodio backend (not ALSA Direct), not streaming, not a user skip.
Falls back to normal stop-play flow when ineligible.
Replace static ReplayGain-only normalization with real-time loudness
analysis that works with any audio source including Qobuz streams.
New components:
- DynamicAmplify: Source wrapper that reads gain from shared atomic
with 50ms linear ramp to prevent clicks on gain changes
- AnalyzerTap: Source wrapper that batches samples and sends them
to the analyzer thread via bounded channel (never blocks audio)
- LoudnessAnalyzer: Background thread computing EBU R128 integrated
LUFS, first measurement at ~10s, refinement every ~5s
- LoudnessCache: SQLite persistence for computed LUFS values so
replayed tracks get instant gain application
Pipeline when normalization ON:
Decode -> Diagnostic -> AnalyzerTap -> DynamicAmplify -> Visualizer
Pipeline when normalization OFF (bit-perfect, unchanged):
Decode -> Diagnostic -> Visualizer
Re-enable normalization toggle in NowPlayingBar UI.
Add normalization_gain field to SharedState and PlaybackEvent so the
frontend can display whether normalization is actively applied to the
current track. The gain factor is stored as AtomicU32 (f32 bits) and
emitted with each playback:state event.
cc
Add ReplayGain-based volume normalization to the audio pipeline.
When enabled, extracts ReplayGain track gain/peak tags from audio
file metadata (Vorbis comments, ID3v2 TXXX) and applies a gain
factor via rodio's Amplify wrapper.
Pipeline order: Diagnostic (raw) -> Amplify (normalization) -> Visualizer
When normalization is OFF (default), no Amplify wrapper is inserted
and the pipeline remains 100% bit-perfect.
- New module: audio/loudness.rs (extraction + gain calculation)
- New settings: normalization_enabled, normalization_target_lufs
- New Tauri commands: set_audio_normalization_enabled/target
- Supports -14 LUFS (default), -18 LUFS, -23 LUFS targets
- Clipping prevention using peak metadata
- Gain preserved across resume/seek operations
cc
Lock-free OR-mask analysis of f32 samples converted to i32,
bypassing PipeWire processing entirely. DiagnosticSource wraps
the audio source at decode level for both rodio and ALSA Direct
paths. Tauri commands: start_bitdepth_capture / stop_bitdepth_capture.
cc
ReinitDevice was resetting position to 0 and clearing audio data,
making it impossible to resume after changing audio settings.
Now uses pause_playback_timer() to capture real-time position and
keeps audio data intact. The existing Resume handler already
recreates the engine, re-decodes, and seeks to the saved position.
Removes unnecessary stop_playback calls from frontend that were
destroying the preserved state after backend/device changes.
PlaybackEngine::append() now accepts Source<Item = f32>.
ALSA Direct thread uses f32 buffer and calls write_f32().
New write_f32() method in AlsaDirectStream converts f32 samples to
hardware format with full 24-bit precision:
- S24_3LE: f32 -> 24-bit packed 3 bytes (SMSL/Topping DACs)
- S32LE: f32 -> i32 full range
- S24LE: f32 -> 24-bit in i32 container
- S16LE: f32 -> i16
- FloatLE: direct write
Old write(&[i16]) kept for safety during migration.
Update AudioSpecs, decode_with_symphonia(), decode_with_fallback(),
wrap_source closure, and all Source boxing casts to use f32 samples.
This ensures 24-bit audio data flows through the player without
truncation to 16-bit.
- Add shuffle/repeat fields to PlaybackEvent struct
- Emit queue:state event when shuffle/repeat changes via API
- Add queue event listener in desktop frontend
- Include shuffle/repeat in WebSocket broadcasts
- PWA updates shuffle/repeat from WebSocket messages
This enables bidirectional sync: changes from PWA update desktop UI,
and changes from desktop update PWA via WebSocket.
- Add AlsaDirectError enum for structured error classification
- Add BitPerfectMode enum to track runtime bit-perfect status
- Implement automatic fallback: hw fails with format error → try plughw
- Abort on non-recoverable errors (busy, permissions, invalid params)
- Respect ALSA plugin selection from UI settings
- Log runtime mode for debugging
This allows SMSL-class DACs that fail on hw: to work via plughw:
while maintaining bit-perfect guarantees (format conversion only,
no resampling).
cc
- Add VisualizerTap to Player for sample capture
- Create wrap_source helper in audio thread
- Wrap all audio sources with TappedSource at 4 append points
- Player now captures samples for visualization when enabled
- Zero overhead when disabled (fast atomic check)
Audio samples flow: Source -> TappedSource -> RingBuffer -> FFT Thread -> Frontend
1. play_track now returns PlayTrackResult with format_id from Qobuz API
- Returns actual format (5=MP3, 6=FLAC16, 7=FLAC24, 27=Hi-Res)
- Returns None for cache hits (format unknown)
2. Frontend updates track.format based on format_id
- QualityBadge can now correctly show "MP3" when format_id=5
- Previously badge showed "CD" even for MP3 because format was empty
3. Device name now set after backend system init
- PipeWire and ALSA backends now set current_device from settings
- Fixes tooltip showing no device when using backend system
Issue #34 - Display actual stream quality, not track metadata
Backend changes:
- Add sample_rate and bit_depth to PlaybackEvent
- Add stream quality tracking to SharedState
- Extract bit_depth from FLAC/M4A headers in streaming path
- Player sets stream quality when decoding audio
Frontend changes:
- Update PlaybackEvent interface with new fields
- Update currentTrack with actual stream quality from backend
This allows the UI to show the actual quality being played (e.g., 44.1kHz
when streaming a 192kHz track at CD Quality) instead of the track's
maximum advertised quality.
Previously, streaming tracks could not be resumed after pause timeout
because current_audio_data was never set for streaming playback.
Changes:
- Store BufferedMediaSource in current_streaming_source for streaming tracks
- On resume, check if streaming source has complete data and use it
- Clear streaming source on Stop and when switching to non-streaming play
- Log buffered bytes when streaming not complete yet
This allows users to pause streaming tracks and resume them after the
download completes, instead of getting "no audio data available" error.
The playback thread was terminating on pause instead of waiting.
Now uses two separate flags:
- is_playing: controls pause/resume (thread waits when false)
- should_stop: controls full termination (thread exits when true)
This allows the ALSA Direct playback thread to pause and resume
without losing its position or requiring a full restart.
Bit-perfect audio path remains unchanged - only control flow modified.
- Pass track duration from frontend to streaming playback for seekbar
- Add StreamInfo struct with speed measurement during probe
- Implement StreamingConfig::from_speed_mbps() for adaptive buffer sizing
- Add play_streaming_dynamic() method using speed-based buffer config
- Enhanced download logging with progress %, speed (current/avg), and ETA
- Auto-cache streamed tracks for instant replay
The seekbar now shows playback progress during streaming by using
the track's known duration from metadata rather than detecting it
from incomplete audio data.
BREAKING CHANGE: Replace blocking decode_streaming_source with IncrementalStreamingSource
Key improvements:
- IncrementalStreamingSource implements rodio::Source and decodes packets on-demand
- Playback starts immediately after initial buffer (512KB default) fills
- Audio is decoded incrementally while download continues in background
- Reduced initial buffer from 2MB to 512KB for faster startup
- Added fast_start() config option for minimal 256KB buffer
The previous implementation blocked until the ENTIRE file was downloaded
and decoded, making streaming slower than traditional download-first.
Now playback truly starts within ~1 second of initial buffer filling.
Signed-off-by: cpvsc
- Add PlayStreaming variant to AudioCommand enum
- Add play_streaming() method to Player for starting stream-first playback
- Modify play_track command to check stream_first_track setting
- Add get_stream_info() for HEAD request to get content-length
- Add extract_audio_format_from_header() to probe audio format from initial bytes
- Add download_and_stream() for async chunked download with buffer push
- Update BufferedMediaSource with create_reader() for independent read positions
- Add has_min_buffer() method for playback readiness check
- Implement Clone for BufferWriter
Phases 3-4/7: Conditional streaming path with async download integration
cpvsc
- Create streaming_source.rs with BufferedMediaSource implementation
- Implements Read + Seek + MediaSource traits for symphonia compatibility
- BufferWriter allows async chunk pushing from HTTP downloads
- StreamingConfig with configurable min_buffer and max_buffer sizes
- Supports seeking within buffered range and EOF detection
- Thread-safe with Arc<Mutex<>> interior for shared state
- Includes comprehensive unit test suite
Phase 2/7: Core streaming infrastructure for first-track playback
cpvsc
Add 50ms delay after stopping the previous playback engine to allow
the rodio Sink to fully release its reference to the OutputStreamHandle.
This prevents "resource busy" errors when rapidly switching tracks,
especially when selecting from search while radio is playing.
Removed special case that forced ALAC to use symphonia fallback which
decoded the entire file (~30MB) before playback, causing 6+ second delays.
Now ALAC uses rodio's Decoder first (supports ALAC via symphonia-all
feature) which streams instead of decoding everything upfront.
Fallback chain is now:
1. Rodio Decoder (streaming) - instant
2. MP4 fallback (streaming) - instant
3. Symphonia (full decode) - slow, last resort
This brings ALAC playback start time from ~6 seconds to instant,
matching FLAC and other formats.
ALAC files (and some other formats) don't include channel information in
the initial codec params, causing playback to fail with error:
'No channel info in codec params'
Changed to use .unwrap_or(2) instead of .ok_or_else() to default to
stereo (2 channels) which is the most common case and matches the
behavior in metadata extraction.
Fixes playback of ALAC (.m4a) files in Local Library.
This fixes the 2-5s delay when rapidly changing tracks. The drain() was
blocking for hundreds of milliseconds waiting for the PCM buffer to empty,
even when the user was skipping tracks.
Root cause (playback_engine.rs:136):
- drain() was called ALWAYS after playback loop ended
- drain() blocks until PCM buffer is completely empty (~125-500ms)
- When user skips tracks, we don't need elegant buffer drain
- This caused 'need to wait 2-5s' or 'double click next' behavior
Solution:
- Added 'natural_end' flag to track WHY playback stopped
- natural_end=true: Song finished normally → DO drain (elegant end)
- natural_end=false: User skipped/paused → SKIP drain (instant response)
Benefits:
- Track changes now instant (no 2-5s wait)
- Songs that finish naturally still drain properly (no clicks/pops)
- Maintains audio quality while improving responsiveness
Testing:
- Rapid track skipping should be instant
- Songs that play to end should have smooth transition
- No audio artifacts (clicks/pops) in either case
This addresses the 'slow playback' issue on first large sample rate transition
(e.g., 88.2kHz → 44.1kHz). The hardware needs extra time to stabilize when
changing sample rates significantly, especially on cold start.
Root cause:
- First transition 88.2→44.1 kHz: song started slowly
- Subsequent identical transitions: clean
- Affects both ALSA Direct and PipeWire DAC Passthrough
- Hardware 'warm-up' time needed for large rate changes
Fix:
- Increased delay from 50ms to 150ms (still imperceptible to user)
- Gives DAC hardware more time to stabilize clock after rate change
- Extra time especially critical for downward transitions (88.2→44.1)
Testing needed:
- Verify first 88.2→44.1 transition is now clean
- Ensure 150ms delay doesn't create noticeable gap
- Check all other transitions still work
The badge was always showing 'Default' regardless of the selected device because
thread_state.set_current_device() was only being called in the legacy CPAL path,
not in the backend system path (PipeWire DAC, ALSA Direct).
Root cause:
- Lines 801-805: Backend system creates stream but never sets current_device
- Lines 847, 883: Only legacy CPAL path was setting device name from CPAL device.name()
- UI badge calls get_audio_output_status() which returns player.state.current_device()
- current_device was None/outdated when using backend system
Fix:
- After successful stream creation (line 897), read output_device from settings
- Set thread_state.current_device with the device name from settings
- Fallback to 'Default' string if output_device is None
- Now badge shows: 'hw:4,0', 'DacMagic Plus Analog Stereo', etc.
Testing:
- ALSA Direct: Should show 'hw:4,0' or configured device
- PipeWire DAC: Should show 'DacMagic Plus Analog Stereo' or selected sink
- System Default: Should show 'Default'
This is an integral fix that resolves the multi-second gap when changing
sample rates, without breaking bit-perfect modes (DAC Passthrough, ALSA Direct).
Root causes identified:
1. Missing ALSA Direct format change detection - stream wasn't recreated
2. Unnecessary CPAL device enumeration before backend system - added latency
3. Decoder started before stream fully initialized - seekbar advanced without audio
Changes made:
1. Added ALSA Direct to format change detection (line 773-775)
- Stream now recreates on sample rate change for ALSA Direct
- Maintains bit-perfect playback like DAC Passthrough
2. Restructured device initialization to skip CPAL enumeration (lines 798-893)
- Backend system (PipeWire DAC, ALSA Direct) checked FIRST
- CPAL enumeration only used as fallback for legacy path
- Eliminates "Device 'hw:4,0' not found" warning for ALSA Direct
3. Added 50ms delay after stream creation (line 904-906)
- Ensures stream is fully initialized before decoder starts
- Prevents seekbar advancing without audio output
- Small enough not to be noticeable, large enough to prevent gap
Testing needed:
- ALSA Direct: 192→96→192→44.1→48→88.2 kHz transitions
- PipeWire + DAC Passthrough: Same sample rate transitions
- System Default: Verify gapless still works
- Verify bit-perfect modes still work (no resampling)
- Add hardware_volume parameter to PlaybackEngine::new_alsa_direct()
- Update all three call sites in player.rs to pass alsa_hardware_volume setting
- Implement conditional volume control: hardware mixer or DAC/amplifier
- Add device_id() getter to AlsaDirectStream for mixer access
- Hardware volume disabled by default for maximum compatibility
PROBLEM: Hardcoding S32_LE only worked for Cambridge Audio DAC, not generic.
SOLUTION: Auto-detect best format at runtime:
- Try formats in priority order: FloatLE > S32LE > S24LE > S16LE
- Store detected format in AlsaDirectStream struct
- write() auto-converts from i16 based on detected format:
* FloatLE: i16 -> f32 (normalize to -1.0..1.0)
* S32LE: i16 -> i32 (bit-perfect: shift left 16 bits)
* S16LE: i16 -> i16 (no conversion)
Now works with any DAC regardless of supported formats.
PROBLEM: Cambridge Audio DAC only supports S32_LE format, not FloatLE.
Previous code failed with 'Invalid argument (22)' when trying to set FloatLE.
SOLUTION:
- alsa_direct.rs: Changed format from FloatLE to S32_LE
- alsa_direct.rs: Changed write() to accept i32 samples instead of f32
- playback_engine.rs: Convert i16 to i32 using bit-perfect left shift:
i32 = (i16 as i32) << 16
This maintains full bit-perfect precision while using hardware-supported format.
PROBLEM: Original code tried to collect() entire audio source into memory
before playback, causing 600+ MB allocations for 192kHz tracks and silent
failures.
SOLUTION: Stream samples in 8192-frame chunks:
- Read chunk from source iterator
- Convert i16 to f32
- Write to ALSA immediately
- Repeat until end of stream
This reduces memory usage from 600+ MB to ~300 KB per chunk and fixes
the no-audio issue.
- Replace Sink with PlaybackEngine abstraction throughout player
- Update StreamType to use Arc<AlsaDirectStream> for sharing
- Adapt all player commands (Play, Pause, Resume, Seek, Stop, SetVolume, ReinitDevice)
- Add as_any() trait method for backend downcasting
- Fix all type mismatches and compilation errors
- Compilation successful with 0 errors, 30 warnings
Previously, init_device() used hardcoded 48000Hz when reinitializing
streams with backend system. This broke bit-perfect playback after
pause/resume or device changes.
Now init_device() accepts sample_rate and channels parameters and uses
the last known values from current_sample_rate/current_channels to
maintain the correct format across reinitializations.
This ensures DAC Passthrough works correctly:
- 192kHz track stays at 192kHz through pause/resume
- No unwanted resampling to 48kHz
- Bit-perfect playback maintained
Fixes apply to:
- Auto-reinitialization after sink failures
- Resume command when stream needs recreation
- ReinitDevice command (settings changes)
COMPILATION ERROR FIXED:
- audio_settings field was private
- Can't access from commands module
SOLUTION:
- Added public method: Player::reload_settings(settings)
- Encapsulation preserved
- Commands use public API
VERIFIED: cargo check passes
PROBLEMA ENCONTRADO:
1. Frontend guardaba backend_type correctamente ✅
2. Player leía settings al inicio (backend_type = None) ✅
3. Usuario cambiaba backend en UI → se guardaba en BD ✅
4. Pero Player en memoria tenía settings VIEJOS ❌
5. reinit_audio_device usaba legacy CPAL sin backend ❌
SÍNTOMAS:
- "Backend system not configured, using legacy CPAL path"
- "Device '...' not found, using default"
- Audio salía por laptop speakers, no por DAC
SOLUCIÓN IMPLEMENTADA:
1. init_device() ahora intenta backend system primero:
- Lee thread_settings.backend_type
- Si configurado, llama try_init_stream_with_backend()
- Fallback a legacy CPAL si falla
2. reinit_audio_device() recarga settings desde BD:
- Lee AudioSettings desde database
- Actualiza Player.audio_settings en memoria
- Luego llama reinit_device()
- Garantiza que Player siempre tiene config fresca
AHORA FUNCIONARÁ:
- Cambiar backend → se guarda → se recarga → se usa
- Cambiar device → se guarda → se recarga → se usa
- PULSE_SINK se setea correctamente
- Audio debe salir por Cambridge DAC
TESTING REQUERIDO:
Ir a Settings → Audio → Backend: PipeWire → Device: DacMagic Plus
Reproducir música → debería salir por DAC
PLAYER INTEGRATION:
- Added try_init_stream_with_backend() helper function
- Modified AudioCommand::Play to use backend system when configured
- Falls back to legacy CPAL path if backend_type is None
- Preserves existing behavior for users without backend config
BACKEND PRIORITY:
1. Try backend system (if backend_type is set in audio_settings)
2. Fall back to legacy CPAL direct access (backward compatible)
VERIFIED:
- cargo check passes
- Backward compatible with existing installations
- Ready for Phase 5 (UI components)
Add auto-disable logic to prevent incompatible settings from being enabled simultaneously:
Frontend (SettingsView.svelte):
- DAC Passthrough ON → auto-disables Gapless Playback and Crossfade
- Gapless Playback ON → auto-disables DAC Passthrough
- Crossfade > 0 → auto-disables DAC Passthrough
- Added handleGaplessPlaybackChange() and handleCrossfadeChange() handlers
- Modified handleDacPassthroughChange() to disable incompatible settings
Backend (player/mod.rs):
- Only recreate OutputStream on format change when DAC passthrough is enabled
- When DAC passthrough is OFF: reuse existing stream for gapless transitions
- When DAC passthrough is ON: recreate stream for bit-perfect playback
- Added detailed logging for stream recreation decisions
This fixes the issue where Gapless Playback was enabled but gaps still occurred
due to DAC passthrough forcing stream recreation on sample rate changes.
The play_data function was calling decode_with_symphonia which decoded
the entire audio file (271MB+) just to extract sample_rate and channels.
This caused 30-35 second delays between download complete and playback.
New extract_audio_metadata function:
- For FLAC: uses rodio decoder (reads headers only)
- For isomp4: uses symphonia probe to read codec_params without decoding
This restores the fast playback we had before.
This commit implements proper sample rate passthrough for high-quality
audio playback, addressing issue #10 where DAC sample rates were locked
at 44.1kHz regardless of file quality.
Core Changes:
- Modified Player to accept and store AudioSettings (exclusive mode, DAC passthrough)
- Updated AudioCommand::Play to include sample_rate and channels metadata
- Created create_output_stream_with_config() to build custom StreamConfig
- Added sample rate tracking to detect changes between tracks
- Modified Play handler to recreate OutputStream when sample rate changes
- Updated play_data() to extract audio metadata before playback
- Modified lib.rs to read AudioSettings on startup and pass to Player
Device Detection Improvements (Issue #4):
- Added display_name and device_type fields to AudioDevice struct
- Implemented get_pipewire_device_descriptions() to extract friendly names
- Updated get_audio_devices() to map technical names to user-friendly names
- Added device type detection (USB, HDMI, PCI, Bluetooth)
Technical Details:
- Audio files are decoded twice: once for metadata, once for playback
- Stream is recreated only when sample rate or channel count changes
- Falls back to default device config if custom config fails
- Supports both PipeWire and PulseAudio device naming via pactl
- Logs sample rate transitions and OutputStream configuration
Files Modified:
- src-tauri/src/player/mod.rs: Core playback engine changes
- src-tauri/src/lib.rs: Initialize Player with AudioSettings
- src-tauri/src/commands/playback.rs: Device enumeration improvements
Tested: Compiles successfully on Linux with Rust 1.92