−189.1dB
Polyphase upsampling stopband rejection
Measured at 44.1k → 384k. 52 named filter presets built in, 0.0000 dB passband ripple, worst harmonic −134.54 dBc.
!DOCTYPE html>
RLPlayer v1.1.4 · macOS Native Hi-Fi Audio Engine
Combines 52 polyphase rational Sinc upsamplers, four families of high-order ΔΣ modulators, zero-latency uniformly partitioned FFT convolution, 20-band RBJ parametric EQ, and CoreAudio Hog physical exclusive mode in a single native process. The entire chain is an in-house C++ algorithm, and every acoustic metric can be independently recomputed and verified.
Objective Benchmarks
Measured on a real Apple M4 machine with the 132-test automated suite, covering polyphase resampler stopband attenuation, ΔΣ modulation in-band SNR, partitioned convolution mathematical accuracy, and end-to-end bitstream consistency.
−189.1dB
Measured at 44.1k → 384k. 52 named filter presets built in, 0.0000 dB passband ripple, worst harmonic −134.54 dBc.
113.5dB
Measured at DSD256. Covers four modulator families: Butterworth, 7th-order Chebyshev, EC error feedback, and AHM multilevel.
1,058,400bytes
At the same rate with no DSP engaged, the original 1-bit stream is passed through and wrapped as DoP; a million-byte bit-level comparison shows zero deviation.
≤ 6.1e−14
Zero-latency uniformly partitioned overlap-save convolution versus direct convolution, reaching the accuracy limit of double-precision floating-point arithmetic.
20bands
ISO 1/3-octave center-frequency grid, ±24 dB range. A single +6 dB band measures 6.0000 dB at 44.1 / 48 / 96 / 192 kHz.
6,230+
The complete OPRA open acoustic database is built in, searchable by brand and model in milliseconds, and loaded into the parametric EQ in one click.
19
Scans THD+N, noise floor, dynamic range, LRA, crest factor, channel correlation, and high-frequency rolloff cut-off to identify lossy transcoded files.
LNS15/ 5th–15th-order psychoacoustic noise shaping
Five noise-shaping stages (NS1 / NS4 / NS5 / NS9 / LNS15) and four dither types (RPDF / TPDF / Gauss1 / Shaped); in-band SNR gain reaches +86.5 dB from 44.1k → 705.6k.
132
automated regression tests
covering DSP precision, 0 dBFS full-scale overload recovery, and bitstream consistency
Benchmark: measured and reviewed on Apple M4 · macOS with the full CTest automated gate and the offline signal--analyzeanalyzer ().
Signal Architecture
From audio container byte parsing, DirectSDM bitstream passthrough, and the double-precision floating-point DSP pipeline to CoreAudio physical exclusive mode and LAN lossless casting, every stage's state and timing can be inspected in real time.
01
Native C++ parsing.dsfof (LSB-first block interleaved).dffand (MSB-first contiguous frames) 1-bit audio streams; FLAC / WAV / MP3 / OGG use in-house header-reading parsers, with tag scanning 60× to 250×ffprobefaster than external calls.
02
When the DSD source matches the target output rate and no EQ, convolution, or gain adjustment is engaged, the original 1-bit stream skips float conversion and is wrapped directly as a DoP bitstream; measured 1,058,400 bytes bit-identical end to end.
Six-state Bit-Perfect badge:The bottom bar shows "passthrough / PCM→PCM / PCM→DSD / DSD→DSD × bit-exact or not" in real time; click to open the full signal-chain insight panel.
03
Once in the processing pipeline, the audio stream passes in order through loudness and oversampling peak protection, 20-band parametric EQ, zero-latency partitioned FFT convolution, polyphase rational Sinc resampling, and output noise shaping / ΔΣ modulation.
float64computation with 128-sample crossfade hot updates) and zero-latency uniformly partitioned FFT convolution (with Apple Metal GPU acceleration).04
Hog Mode exclusive accessAUHALthrough the macOS low-level unit locks the hardware's physical sample rate and bit depth directly, bypassing system mixer resampling entirely; UPnP streaming upsampling sources and in-app AirPlay push are also integrated natively.
DeviceBusy, and triggers a safe shutdown within 2 seconds when a USB DAC is unplugged or anDEVICE_LOSTendpoint drops.0x05 / 0xFA, with legal phase-continuous DoP silence frames filled automatically across partitions, pause/resume, and track ends.AudioToolbox(44.1 kHz / 16-bit) real-time lossless encoding, and RTP clock synchronization.
Available in v1.1.4
Combines mastering-grade local playback, multi-protocol LAN lossless casting, driver-free system capture, headphone acoustic calibration, audio quality assessment, and classical library management.
Takes over Hog Mode through the CoreAudioAUHALlow-level unit and locks the hardware's physical stream format (), automaticallykAudioStreamPropertyPhysicalFormatprobing every sample rate and bit depth the current DAC supports.
CoreAudio AUHAL · Hog Mode · Physical Format Lock
Native DSF / DFF container parsing, with DirectSDM 1-bit bit-perfect passthrough, same-family integer-ratio DSD64 ↔ 128 ↔ 256 pure-DSD-domain rate conversion, and three high-precision decimators (FIR / IIR / CIC).
1,058,400bytes bit-identical end to end
SSDP discovers LAN streamer renderers across multiple network interfaces (automatically bypassing proxy virtual adapters), and supports direct HTTP Range 206 serving of original files, or partitioned streaming upsampling through the local 52 polyphase filters before pushing high-sample-rate WAV.
Built on Apple and the privateCATapDescriptionAudioHardwareCreateProcessTapaggregate device mechanism, it captures the system mix losslessly in the background with no third-party virtual sound card such as BlackHole and no screen-recording permission.
No kernel-driver intrusion:Implemented entirely on the native macOS 14.2+ CoreAudio Process Tap API; ready to use right away.
The complete Open Product Research Audio (OPRA) database is built in, searchable by brand and model in milliseconds, with explanations of 11 target tuning curve families such as Harman plus SVG frequency-response previews, applied to the 20-band parametric EQ in one click.
⌥ Optionfrequency horizontally against slips) with automatic negative pre-gain anti-clipping.With one click, scans a track's THD+N, noise floor, dynamic range (DR), loudness range (LRA), crest factor, left/right channel correlation, mono fold-down loss, and high-frequency rolloff cut-off to pinpoint "fake lossless" tracks that are actually lossy upsampled files.
Nativelybuilds the four-level classical entity model COMPOSER → Work → Movement → Performance, with real-time Discogsglobal recording lookup by work name and composer (conductor / orchestra / soloist / label / era) and intelligent de-duplication by performance.
[mm:ss.xxx]lyrics with timecode./api/llm/modelsmodel discovery via ).
Uses a pure Cocoa single-processNSWindow + WKWebViewhost; front end and back end communicate directlyrlplayer://app/in memory through the in-process custom scheme, and the main process listens on no local network port.
librtagpu.dylib: compiles Metal Shading Language (MSL) kernels at runtime to accelerate polyphase Sinc upsampling and long-impulse-response FFT convolution.Shiftfine adjustment (double-click to reset to zero).Apple Silicon arm64 · Single-Process In-Memory IPC · Zero Open Ports
Engineering Stack
No dependency on Electron, Python, or external background service processes. From the macOS window host and digital signal processing operators to the LAN streaming protocols, everything is natively built with C++17 and Objective-C++.
01 · HOST & RUNTIME
The host layertools/mac_host.mm() uses native andAppKitWebKitand statically linkslibrlapp_corethe core engine into the same process. A interceptsWKURLSchemeHandlerrequestsrlplayer://app/and dispatches binary-safe responses on GCD concurrent queuesNSData; a channel and aWKScriptMessageHandlerWithReplybundled self-compiled slim () complete thearm64picture, soFFmpeg 8.1.2(ffmpeg / ffprobeusers need no separate command-line decoding environment.
02 · DSP & METAL KERNELS
Polyphase rational Sinc resamplingresampler.h(), uniformly partitioned overlap-save FFT convolution (), 5th–15th-orderconvolution.hnoise shaping ()dither.h, and four families of ΔΣ modulatorsmodulator.h() all run in 64-bitdoubleprecision; the 20-band RBJ biquad PEQ () uses coefficientpeq.hcomputationfloat64with 128-sample crossfade hot updateslibrtagpu.dylib; compiles Metal Shading Language kernels on demand to accelerate heavy workloads.
03 · AUDIO I/O & PROTOCOLS
Audio output drives , , andAudioUnit (AUHAL)、kAudioDevicePropertyHogModedirectlykAudioStreamPropertyPhysicalFormat; system loopback is built on macOS 14.2+ and privateCATapDescriptionAudioHardwareCreateProcessTapaggregate devices; network casting and the data center are implementedlibcurl、AudioToolboxon , (real-time ALAC encoding), nativeSSDP、mDNS (_raop._tcp)RTP, , and the stack.
Roadmap & Research
Building on the existing in-house signal chain and native Apple Silicon architecture, the roadmap keeps moving toward user-space USB Audio 2.0 hardware passthrough, multichannel room acoustic measurement, reference-grade DSP 2.0, and sample-accurate gapless playback.
Builds a complete macOS user-space USB Audio Class 2.0 (UAC2) protocol stack inside the application process: it takes over the USB DAC audioIOUSBHost / IOKitofficially public IOUSBHost / IOKit interfaces, bypassing CoreAudio's PCM barrier entirely. No kernel extensionextension is required (no kext), and SIP (System Integrity Protection) need not be disabled, and the interface is released smoothly back to the system when playback stops or the app quits.
CS_INTERFACE, independent clock domains, and DSD AltSettings, compatible with native bit packing on XMOS, Thesycon, Comtrue (CT7601), and Amanero ().DSD_U32_BE / LE / U16)。Breaking past macOS audio limits:Eliminates the 33% USB bandwidth waste and the halved sample rate imposed by DoP wrapping, fully unlocking hardware native DSD512 / DSD1024 on flagship DACs whose PCM ceiling is 768 kHz.
Closes the real-time loop of "Process Tap system capture / physical capture → 64-bit in-house DSP → exclusive output", letting the whole Mac's streaming audio, video soundtracks, and even an external phono preamp input flow in real time through the 20-band PEQ, OPRA headphone calibration, room FIR convolution, and polyphase upsamplers.
room_correction(theCoreAudioRoomTransportunderlying and are already in place): Farina logarithmic sweep → synchronized measurement-mic capture → frequency-domain deconvolution → psychoacoustic smoothing → automatic generation of a 32-bit Float minimum-phase correction FIR per channel.2.0 / 2.1 / 5.1 / 7.1 / 5.1.2 / 7.1.4six speaker layouts (, including 80 Hz LFE bass management), with one-click A/B blind listening and automatic mounting by output layout.Upgrades the DSD pipeline output architecture: after PEQ, convolution, and polyphase upsampling, the 64-bit double-precision floating-point planar data no longer passes through an intermediate 24-bit PCM quantization step and goes straight to the ΔΣ modulator, with modulation headroom computed dynamically from the true interpolated signal envelope.
poly-sinc-ext2、poly-sinc-xtr、sinc-MGa / Lh、minringFIR-lpsuch as , , , and ) with adaptive tap-budget back-solving.vDSPnew runtime compute backends and a floating-point accuracy self-verification report (CPU FP64 / Metal FP32).Embeds an L0 / L1 / L2 three-level collapsible real-time analyzer in the now-playing page (with per-track de-duplication, controlled concurrency, and immediate cancellation on track change), and introduces independent 4× / 8× oversampled true peak (True Peak dBTP) measurement.
Upgrades the Server-Sent Events (SSE) streaming parser to show the progress of "authoritative encyclopedia fetch → release verification → in-depth guide generation" in real time and render content with a streaming typewriter effect; adds Album Context and citation attribution checks.
Front Cover, Type 3file construction → bit-level PCM stream comparison and MP3 Gapless Delay boundary validation → original file backup → atomic replacement → lossless rollback on error.Builds background pre-decode slots across tracks and seamless bridging of DSP filter history state, giving live concert, opera, and classical symphony albums a 0-sample silent gap when tracks change.
0x05 / 0xFAcontinuous) and a 10–20 ms cosine-smoothed volume ramp (automatic bypass at 0 dB) eliminate track-change pops and gain-adjustment zipper noise.MPNowPlayingInfoCenter / MPRemoteCommandCenterCenter (), multi-display window coordinate memory, cover cache LRU capacity management, and Apple Developer ID signing and notarization.Starting from the in-house DSP foundation already shipped, it advances step by step along clear technical dependencies and engineering gates.
52 polyphase Sinc upsamplers, 4 SDM families, DirectSDM passthrough, CoreAudio Hog exclusive mode, UPnP streaming upsampling, in-app AirPlay, 6,230+ OPRA, the four-level classical model, and three-level real-time analysis / dBTP true peak on the track page.
SSE streaming AI guides, library-wide cover fill workbench, five-gate cover embedding, and macOS platform QoS polish.
2.0–7.1.4 room acoustics measurement wizard, system-wide real-time DSP rack, 79+ named filter matrix, 64-bit Float DSD modulation pipeline, and multi-stage cascaded interpolation.
In-house user-space UAC2 USB asynchronous driver stack (true native DSD64–1024 passthrough and upsampled direct output) and a sample-accurate zero-gap gapless playback engine.
Download & Open Source
Native Apple Silicon (arm64) standalone installer with all decoding and Metal compute components built in; drag it into Applications and it runs.
v1.1.4Latest stable release
macOS 13.0+ · native Apple Silicon arm64. The app bundle already includes the self-compiled decoding components, (Metal kernels)ffmpeg / ffprobe, and offlinelibrtagpu.dylibdiagnostic tools, so no external dependency environment needs to be set up.
34b9305bf597228c2181f7e1ca3fc29675c8a774b88e976442acc1f0c7568761
verification commandshasum -a 256 RLPlayer-1.1.4-macOS-arm64.dmg
xattr -cr /Applications/RLPlayer.app
Applications, run this once in Terminal to clear the quarantine attribute and open it normally.
Credits & Community
Thanks to everyone who took part in RLPlayer's development, visual design, real-device testing, and community building.
Feature R&D collaboration, engineering integration, and module improvement
Brand visual identity, product icons, and media art design
Multi-device real-hardware listening tests, edge-case investigation, and feedback