Go backend:
- Add AudioCodec field to DownloadResult and DownloadResponse
- Extension download results can now include audio_codec/audioCodec
- ffmpegGetInfo and probeAudioQuality now return codec field
- Add trackItemBytes option to file.download() for custom progress handling
Flutter:
- Check audio_codec before container conversion
- Skip FLAC conversion if source codec is lossy (AAC, MP3, Opus, etc.)
- Prevents fake upscale from lossy to lossless containers
- Extract _formatDownloadProgressLabel() for cleaner code
- Show received/total size when bytesTotal is available
- Estimate total size from progress when only bytesReceived is known
- Add text overflow handling with ellipsis
New lyrics providers using Paxsenix API:
- Spotify: Synced lyrics from Spotify
- Deezer: Synced lyrics from Deezer
- YouTube: Lyrics from YouTube
- Kugou: Lyrics from Kugou (Chinese service)
- Genius: Plain text lyrics from Genius
Implementation:
- Add lyrics client implementations for all providers
- Smart search result scoring based on track name, artist, and duration
- Support for both synced (LRC) and unsynced lyrics formats
- Fallback search with simplified track names and primary artist
UI updates:
- Add provider entries to lyrics priority settings page
- Add display names for new providers in settings
FFmpeg doesn't always embed lyrics correctly to MP3 files. This adds
manual ID3v2.3 USLT (Unsynchronized Lyrics) frame writing after FFmpeg
metadata embedding to ensure lyrics are properly stored.
Implementation:
- Extract lyrics from metadata (UNSYNCEDLYRICS or LYRICS key)
- Build ID3v2.3 compliant USLT frame with UTF-16LE encoding
- Insert or replace USLT frame in existing ID3v2.3 tag
- Create new ID3v2.3 tag if file has no ID3 header
- Skip gracefully for unsupported ID3 versions or flags
Also includes minor audio analysis improvements:
- Consistent dynamic range calculation (peak - rms)
- Filter out 'unknown' and 'n/a' labels
- Add -vn -sn -dn flags for more robust stream selection
Audio Analysis Enhancements:
- Display codec name and container format
- Show decoded sample format (s16, s32, fltp, etc.)
- Add LUFS integrated loudness measurement (broadcast standard)
- Add true peak measurement (dBTP)
- Detect and count clipping samples per channel
- Estimate spectral cutoff frequency (helps detect fake upscales)
- Show per-channel statistics (Peak, RMS, DR, Clip count)
UI Improvements:
- MetricChip now handles long text with ellipsis
- Constrained max width for better layout
Cache version bumped to 4 to force rescan with new metrics.
- Replace inline format detection with convertibleAudioSourceFormat()
- Replace inline conversion rules with canConvertAudioFormat()
- Add unit tests for Dolby format detection and conversion rules
Audio Analysis:
- Add rescan capability by bumping cache version
- Display channel layout (stereo, 5.1, etc.) and bitrate
- Use astats filter for more accurate peak/RMS measurements
- Support more formats: mp4, ac3, eac3, mka, wv, ape, tta, aif
- Only report bit depth for codecs that store it (FLAC, ALAC, WAV)
- Validate cache for SAF content:// URIs
Conversion:
- Add AAC as conversion target format
- Recognize ALAC as lossless source
- Prevent accidental deletion when source and target URI match
- Store format and bitrate in database after conversion
Utilities:
- Add audio_conversion_utils.dart for centralized conversion logic
- Add isSameContentUri() helper for safe URI comparison
The HIGH-quality lossy format picker can now produce an AAC/M4A 320 kbps output alongside MP3 and Opus. FFmpegService.convertM4aToLossy/convertAudioFormat, the Dart queue pipeline, the Kotlin finalizer, and the library database format helper all route .m4a through a unified aac codec path and tag the resulting file with the M4A metadata writer. The Lossy Format setting gains a new option, and the track metadata convert dialog lists AAC next to the other targets.
Apple Music lyrics gain a 'eLRC word sync' switch (default off). When disabled the pax-to-LRC formatter strips inline word timestamps, producing line-synced LRC that is safer for players that choke on eLRC; enabling it restores the previous word-by-word behaviour. The change propagates through SetLyricsFetchOptions and invalidates the global lyrics cache on toggle.
Broad l10n migration: roughly 400 previously hardcoded English strings across queue, settings, track metadata, repo, audio analysis, setup and extension screens now live in the ARB catalog, with matching plural/placeholder forms. No behaviour change beyond localisation. Existing and new unit tests (lyrics eLRC toggle and Dart settings round-trip) pass.
Introduce AppRemoteConfigService which fetches a platform/version/locale-aware JSON payload from api.zarz.moe/v1/spotiflac-mobile/config and caches it in SharedPreferences. main_shell shows a one-shot announcement dialog (respecting dismissible, CTA, time window and version gates) when no update prompt is pending; dismissed IDs are persisted so each announcement surfaces only once.
Tweaks bundled in: the service health dot loses its blur halo in favour of solid Material 3 tones, and AppInfo gains the remote config endpoint constant. The share listener and SAF migration hook stay synchronous inside the post-frame callback so share-intent URLs never race the network-bound checks.
New unit tests cover the announcement CTA/active-window rules.
The SAF and local post-download branches used to rush an ffmpeg 'M4A to FLAC' remux whenever the output extension was .flac, which silently upscaled AAC or EAC3 streams into a lossless container. Each branch now mirrors the native worker by probing the primary audio codec before converting: lossless sources (and true FLAC-in-MP4 files) stay in their native container with the right extension, while genuine ALAC/WAV payloads still get remuxed.
Add an outputExt field to DownloadRequestPayload so the Go backend always knows the user-requested container, and use it together with _shouldRequestContainerConversion to pick the right behaviour for shouldPreserveNativeM4a and the Kotlin finalizer. Decryption descriptors no longer force M4A preservation on their own; the codec probe already makes that call correctly.
Bump the history schema on both the Kotlin finalizer and the Dart database to v9, adding bitrate (kbps) and format (codec label) columns, and let the download flow fill them from backend/probe metadata so lossy downloads keep a 'AAC 256kbps' label instead of falling back to the stored placeholder. Library filtering and the track metadata screen now read format/bitrate directly from those columns, which also fixes mis-tagged quality badges after re-downloading a track at a different format.
Additional fixes bundled in: EditFileMetadata now routes ReplayGain writes through the M4A path whenever the file starts with ftyp (fixing .flac files that actually hold MP4 containers); GetM4AQuality falls back to the first trak/mdia/mdhd duration when mvhd is zero so EAC3 streams no longer report 0s; and both Kotlin and Dart reject bitrate values below 16 kbps to prevent probe noise from surfacing as '0 kbps' labels. New unit tests cover the EAC3 mdhd fallback and the mis-named M4A replaygain path.
GetM4AQuality now recognizes fLaC, alac, ec-3, ac-3, and ac-4 sample entries and parses the MP4 FLACSpecificBox so library entries carry the real codec rather than the container extension. The AudioQuality struct exposes Codec and Bitrate fields (with an estimator for compressed streams), and ReadFileMetadata publishes format + audio_codec so Flutter and Kotlin can make format decisions based on the actual stream.
Downstream: library_scan labels M4A-family items as flac/alac/eac3/ac3/ac4/m4a, zeroes the bitrate for lossless formats, and the filter UI + quality badges use the codec-derived format instead of only the file extension. Scans and SAF importers also accept .mp4 and .aac file extensions. New unit tests cover codec name mapping and MP4 FLACSpecificBox decoding.
The audio analysis card used to read from a persistent cache but offered no way to refresh the result when the underlying file had been re-downloaded at a different quality (for example, re-downloading a track as FLAC after capturing it as AAC). Add an explicit rescan control that clears the cached JSON + spectrogram, reruns the FFmpeg probe and analysis pipeline, and swaps in the fresh data while keeping the loading copy distinct from first-run analysis. A retry button is also exposed in the error card so transient failures do not require navigating away.
All audio_analysis strings now have a Re-analyze / Re-analyzing pair in the ARB catalog so every locale can translate them independently.
Add a pair of floating quick-scroll buttons on the library tab so long lists become easier to navigate. The buttons sit above the bottom navigation (or the selection toolbar in selection mode), fade in and out based on the active page's scroll metrics, and share their scroll-target keys per filter mode so switching filters does not carry over the previous page's scroll state.
Downloads from providers that stream FLAC inside an fMP4 container (e.g. Amazon Music) were being written to disk with a .flac extension while the payload still carried ISO-BMFF atoms. The container-conversion guard then saw codec=flac and skipped the remux, leaving native FLAC tag writers to fail with 'fLaC head incorrect'.
Force '-f flac' on the decryption command whenever the target extension is .flac so FFmpeg emits a real FLAC stream, and add an 'fLaC' magic-byte probe on both the Dart and Kotlin container-conversion guards so a FLAC-in-MP4 source is remuxed rather than silently passed through as a tag-writer hazard.
GetM4AQuality previously defaulted to 16-bit whenever the audio sample entry was not ALAC, which silently labeled lossy AAC downloads as CD quality in the library and in extension APIs. Only fill BitDepth when the atom is ALAC (including the ALACSpecificConfig refinement), and leave it as zero for AAC/mp4a, matching how the MP3 and Opus probes already report lossy sources. Tests cover both the ALAC and AAC branches.
Zarz Eleutherius