Windows · macOS · iPhone · iPad
HeadMania.org Upsampler plays your library, HIGHRESAUDIO and lossless radio through very long reconstruction filters, then sends the result to your USB DAC and headphones, or over Wi-Fi to any UPnP network streamer in the house. Neuromancer, its optional neural stage, reshapes the sound as the song plays. It is experimental, and it switches on only where your device keeps up in real time.
Also for: macOS (Apple silicon) iPhone & iPad Android remote (test)
Also for: Windows macOS (Apple silicon) Android remote (test)
Free on Windows and Mac. On iPhone and iPad, free below 500,000 taps; the one-time Signature Edition adds the larger filters. You plug in a USB DAC, or stream over Wi-Fi to a UPnP streamer; Diretta works on Windows.
What it does
Your own files and CUE sheets, HIGHRESAUDIO with your subscription, lossless radio, or the sound of any other app.
The engine computes in 64-bit and filters with up to 4 million taps live. You pick the filter's character: Linear, Organik or in between.
To your USB DAC, or over Wi-Fi to any UPnP network streamer in the house; also Diretta, and USB-C DACs on iPhone and iPad. The Play tab shows exactly what is running.
HeadMania is an upsampler, DSP engine and music server for Windows and macOS. It computes in 64-bit floating point, and in double-double (about 106-bit) for Reference Precision. It does the reconstruction filtering your DAC would otherwise do, with filters of up to 4 million taps live and 16 million for file playback. Play your library or lossless radio, filter what other apps play, or run it as a server you control from your phone. The Play tab always shows the filter and rate actually running.
HeadMania turns an iPhone or iPad into a 64-bit upsampler and network renderer. Play your own files or lossless radio, or send music from a UPnP app such as JPLAY or mconnect, then out to a USB-C DAC or a streamer on your network. Filters run up to 1 million taps on iPhone and 4 million on iPad Pro M5 and other M-series iPads. It can also remote-control HeadMania on your computer. The Play tab always shows the filter and rate actually running.
Streaming partner
“A seamless integration inside HeadMania’s UpSampler 3, makes listening to HiRes music, effortlessly, comfortable and more enjoyable for hours.”
HIGHRESAUDIO, the Berlin store and streaming service for high-resolution music, lives in its own HRA Streaming tab. Sign in with your subscription and every track takes the same path as a file on your own disk: your filter, Neuromancer, DSD conversion if you want it, the EQ and the room correction (in testing), then out to your USB DAC or network player.
Needs a HIGHRESAUDIO streaming subscription, where HIGHRESAUDIO offers it.
The engine, in numbers
No slogans: each card says what was measured and what it costs. Pick your platform to see what applies to you.
taps for file and server playback
Computed in double-double, about 106-bit. The step from full level to 100 dB down narrows to 0.14 Hz.
Reference Precision at 4M, 8M and 16M taps. Each doubling halves the cutoff width: 0.57 Hz at 4M, 0.28 Hz at 8M, 0.14 Hz at 16M. Costs seconds of start-up delay and heavy CPU; not for live capture.
taps live on iPhone, and on M-series iPads
The same engine in your pocket, out to a USB-C DAC or a streamer on your network.
Free below 500,000 taps; the one-time Signature Edition adds the larger filters. Over USB-C, iPhone and iPad send PCM, up to ×16; to a UPnP streamer they send DSD too, where it accepts DSD.
delay with Organik Live, up to 1M taps
No block to wait for, so long filters work for video, games and live capture.
Block filters (Live Linear, Live Organik, Cascade) wait 3.7 to 9.8 ms from a 44.1 kHz source. Organik Live filters each sample as it arrives: 0.13 to 0.18 ms. Live capture’s own audio buffers still add tens of milliseconds.
of the drums’ 16–22 kHz attack energy takes Pulse’s fast path
Drum hits get a short filter; everything between them stays on the long one.
Measured on 21 recordings: Touch sends 61%, Pulse 90%. The short path is a 575-tap minimum-phase filter.
learned values in Neuromancer, a neural stage before the filter
Optional. It runs on your graphics card or the Neural Engine, after a speed test on your device.
About 730 billion operations for every second of music. Adds a few seconds of start-up delay; switches on only when your device keeps up. It keeps the graphics card or the Neural Engine busy, which costs power.
DSD from HeadMania’s own modulator
For true 1-bit and hybrid DACs: DSD64 to DSD512, DSD1024 on Windows, and DSD2048 as an experiment. PCM runs from ×2 to ×32.
In-band dynamic range of the modulator, 1 kHz tone, 20 Hz–20 kHz: 224 dB at DSD1024, above 240 dB at DSD2048. Read it as headroom: a DAC’s analog stage typically sits near 120 dB. DSD2048 needs an AVX2 processor.
On Windows and Mac, unzip the download and open the app inside; after that, updates arrive in the app, signed and installed when you choose. All versions and release notes: GitHub releases.
Filters
Every filter family runs from short to very long. Longer filters cut closer to the band edge; the phase choice decides where the ringing goes.
Ringing spreads evenly before and after each transient, with the same delay at every frequency.
6K – 4M tapsThe same frequency response as Linear, with its ringing moved after the transient.
6K – 4M tapsBetween the two: very little ringing before a transient, longer ringing after it than Linear.
6K – 4M tapsOrganik with a fast path for drum attacks.
6K – 4M tapsDouble-double computation at 4M, 8M and 16M taps, for file and server playback.
4M – 16M taps · Windows & MacFor video, games and live capture: no block, under 0.2 ms.
10K – 1M taps · Windows & Mac
Every filter runs in 64-bit floating point (Reference Precision in double-double), with coefficients made for the exact rate ratio. Tap count sets how sharp the cutoff is, not how good the filter is. Figures on this page are software measurements of the filter output, not DAC measurements or listening results.
Linear phase: ringing spreads evenly before and after each transient, with the same delay at every frequency.
Minimum phase: the same frequency response as Linear, with its ringing moved after the transient.
A fixed point between the two (α = 0.25): very little ringing before a transient, longer ringing after it than Linear.
Organik with a fast path for attacks. Touch sends the first 20 ms of each detected hit through a short minimum-phase filter and plays everything else on Organik. Pulse follows longer hits and some percussive sound between them.
Linear, Organik, Organik-Adaptive and Mixed Phase DD at 4M, 8M and 16M taps, for file and server playback. DD means double-double computation, about 106-bit.
For video, games and live capture. Live Linear, Live Organik and Cascade work in blocks: 3–10 ms of delay. Organik Live 10K to 1M has no block: under 0.2 ms.
Pick Linear for a DAC with no filter of its own (NOS, R2R), or when you want the same delay at every frequency. Pick Organik for transient-rich music, or when your DAC applies its own linear-phase filter: two linear-phase filters in a row add their pre-ringing, minimum phase adds almost none. Not sure? Start with Organik 2M.
A sharper cutoff. The step from full level to 100 dB down narrows from 533 Hz at 6K to 25 Hz at 150K, about 2.3 Hz at 1M, 0.57 Hz at 4M and 0.14 Hz at 16M. Each doubling of taps halves it, so the band stays flat closer to 22,050 Hz: half a hertz below the edge, 1M is down 1.82 dB and 4M not at all. Depth is already past what can be measured: from 150K up, images sit about 290 dB down, the analyser's own floor, so sizes cannot be ranked by depth. A typical DAC's analog stage sits near 120 dB. Reference Precision's arithmetic floor, about 633 dB inside and 328 dB at the output, is how precisely the numbers are computed, not what the filter removes.
Each doubling of taps halves the width of the cutoff: 0.57 Hz at 4M, 0.28 Hz at 8M, 0.14 Hz at 16M. For file and server playback; not for live capture.
For video, games and live capture. Live Linear, Live Organik and Cascade work in blocks, and the sound waits for each block: 3.7 to 9.8 ms from a 44.1 kHz source. Organik Live filters each sample as it arrives, so only the filter's own delay remains: 0.13 to 0.18 ms, measured at 1 kHz. More taps buy a steeper edge. Images sit 180 dB down from 22.05 kHz with the 10K and at least 240 dB down with the 100K, 500K and 1M; across those three, the step from flat to 240 dB down narrows from 15 Hz to 3 Hz to 1.4 Hz. Every size is minimum phase. Capture and output buffers add their own delay.
Linear, Organik, Mixed Phase, Touch, Pulse and Organik-Adaptive at 1M–4M run on a faster engine with the same output (difference below −280 dB): up to about 8× faster (Organik 4M, one core). It is the default on Windows, Apple-silicon Macs, iPhone and iPad. The Classic engine stays available in Settings. Fast Start lets the largest filters start sooner; the filter's delay is unchanged.
| Filter | Phase | Taps | THD+N | Image rejection | Stopband floor | To −100 dB | Worst case | Delay | Ring pre / post (ms) |
|---|---|---|---|---|---|---|---|---|---|
| Low latency | |||||||||
| Cascade | Linear | 765 | < −150 | 108.8 | 315.0 | 3,020 Hz | 137.7 | 1.1 ms | 0.64 / 0.64 |
| Cascade Organik | Minimum | 765 | −94.4 | 155.7 | 150.0 | 3,020 Hz | 137.6 | ~0.1 ms | 0.08 / 1.28 |
| Live Linear | Linear | 1.3K | < −150 | 143.1 | 315.6 | 1,813 Hz | 136.2 | 1.8 ms | 0.97 / 0.97 |
| Live Organik | Minimum | 1.3K | −89.0 | 151.1 | 145.5 | 1,814 Hz | 136.3 | ~0.2 ms | 0.08 / 2.09 |
| Core | |||||||||
| Linear 6K | Linear | 5,667 | −126.6 | 202.4 | 343.6 | 533 Hz | 201.6 | 8.0 ms | 2.57 / 2.57 |
| Linear 27K | Linear | 27,267 | −120.4 | 259.0 | 329.9 | 125 Hz | 239.8 | 38 ms | 5.62 / 5.62 |
| Organik 27K | Minimum | 27,267 | −116.7 | 259.0 | 328.7 | 125 Hz | 239.9 | ~0.2 ms | 0.10 / 28.7 |
| Linear 150K | Linear | 149,613 | < −150 | 295.6 | 316.0 | 25 Hz | 274.4 | 212 ms | 7.07 / 7.07 |
| Dark 176K | Linear | 176K | < −150 | 297.6 | 317.1 | 21 Hz | 280.1 | 249 ms | 8.35 / 8.35 |
| Mega | |||||||||
| Linear 1M | Linear | 1,000,001 | −118.8 | 289.4 | 325.9 | ≈ 2.3 Hz † | 272.0 | 1.42 s | 7.20 / 7.20 |
| Organik 1M | Minimum | 1,000,001 | < −150 | 291.1 | 315.3 | ≈ 2.3 Hz † | 272.5 | ~0.2 ms | 0.10 / 102 |
| Mixed Phase 1M | Mixed α 0.25 | 1,000,001 | — | 217.3 | 217.2 | 12.03 Hz | — | 1.42 s | 0.051 / 37.5 |
| Linear 2M | Linear | 2,000,001 | < −150 | 287.2 | 315.6 | ≈ 1.1 Hz † | 277.3 | 2.83 s | 7.20 / 7.20 |
| Organik 2M | Minimum | 2,000,001 | −118.0 | 291.1 | 313.3 | ≈ 1.1 Hz † | 277.3 | ~0.2 ms | 0.10 / 103 |
| Linear 4M | Linear | 4,000,001 | < −150 | 288.8 | 304.1 | 0.57 Hz | 246.8 ‡ | 5.67 s | 7.2 / 7.2 |
| Organik 4M | Minimum | 4,000,001 | < −150 | 288.8 | 300.0 | 0.57 Hz | 246.8 ‡ | ~0 ms | — |
| Organik Touch / Pulse 1M | Min. + short path | 1,000,001 + 575 | Touch: Organik 1M between hits. Short attack path (Pulse also between hits): images 56.6 dB down | +87 ms | path: 0.10 / 2.43 | ||||
| Adaptive and R&D | |||||||||
| Organik-Adaptive 1M | Adaptive min. | 1,000,001 + 575 | < −150 | 291 / 180 | 315 / 305 | ≈ 2.3 † / 3,000 Hz | 272 / 280 | 8.16 ms | 0.10 / 102 · 0.10 / 2.43 |
| Transient-Perfect 1M | Adaptive mono. | 1,000,001 + 22 | < −150 | 291 / 13–97 | 315 / 56 | ≈ 2.3 † / — | 272 / 11.8 | 8.16 ms | 0.10 / 102 · 0.03 / 0.03 |
| Kinetic | Adaptive | 149,613 + 765 | −121.3 | 295.2 | 315.8 | — | — | 212 ms | — |
| Kinetic V2 300K | Adaptive | 300,001 × 3 | < −150 | 295.3 | 316.0 | 13 Hz | 259.9 | 425 ms | 7.18 / 7.18 |
| Kinetic V3 300K | Adaptive | 300,001 × 2 | < −150 | 296.6 | 317.0 | 13 Hz | 280.0 | 433 ms | 7.18 / 7.18 |
| Kinetic V2 1M | Adaptive | 1,000,001 | < −150 | 301.5 | 326.1 | not re-measured | 272.2 | 1.42 s | 7.20 / 7.20 |
| Kinetic V3 1M | Adaptive | 1,000,001 × 2 | −127.1 | 303.2 | 329.4 | not re-measured | 292.3 | 1.43 s | 7.20 / 7.20 |
| Reference Precision (double-double) ♦ | |||||||||
| Linear / Organik 4M | Linear / Min. | 4,000,001 | < −150 | 288.8 | 328 | 0.57 Hz | 246.8 ‡ | 5.67 s / ≈0 | 7.2 / 7.2 · <0.1 / long |
| Kinetic V3 4M | Adaptive | 4,000,001 × 2 | < −150 | 288.8 | 328 | 0.57 Hz | 246.8 ‡ | 5.67 s | 7.2 / 7.2 |
| Organik-Adaptive 4M | Adaptive min. | 4,000,001 + 575 | < −150 | 288.8 | 328 | 0.57 Hz | 246.8 ‡ | ≈0 | <0.1 / long |
| Linear / Organik 8M | Linear / Min. | 8,000,001 | < −150 | 289.4 | 328 | 0.28 Hz | 240.7 ‡ | 11.34 s / ≈0 | 7.2 / 7.2 · <0.1 / long |
| Organik-Adaptive 8M | Adaptive min. | 8,000,001 + 575 | < −150 | 289.4 | 328 | 0.28 Hz | 240.7 ‡ | ≈0 | <0.1 / long |
| Linear / Organik 16M | Linear / Min. | 16,000,001 | < −150 | 288.1 | 328 | 0.14 Hz | 234.7 ‡ | 22.68 s / ≈0 | 7.2 / 7.2 · <0.1 / long |
| Organik-Adaptive 16M | Adaptive min. | 16,000,001 + 575 | < −150 | 288.1 | 328 | 0.14 Hz | 234.7 ‡ | ≈0 | <0.1 / long |
Each Mixed Phase preset is one fixed filter at α=0.25: one convolution path, no transient detector and no crossfade. Mixed Phase DD adds the same choice at 4M, 8M and 16M.
| Preset | Taps | Peak delay | At 352.8 kHz | Dense ×8 | Image ×8 | Impulse pre / post | Step pre / post |
|---|---|---|---|---|---|---|---|
| Mixed Phase 6K | 6,001 | 3,000 samples | 8.50 ms | — | 124.286 dB | 1.732 / 6.117 ms | 0.122 / 4.779 ms |
| Mixed Phase 27K | 27,001 | 13,500 samples | 38.27 ms | 123.42 dB | 123.42 dB | 0.048 / 16.735 ms | 0.048 / 9.430 ms |
| Mixed Phase 150K | 150,001 | 75,000 samples | 212.59 ms | 167.80 dB | 169.31 dB | 0.051 / 30.544 ms | 0.051 / 13.869 ms |
| Mixed Phase Extended 500K | 500,001 | 250,000 samples | 708.62 ms | — | 199.05 dB | — | — |
| Mixed Phase 1M | 1,000,001 | 500,000 samples | 1.417 s | 217.16 dB | 217.25 dB | 0.051 / 37.523 ms | 0.051 / 14.195 ms |
| Mixed Phase 2M | 2,000,001 | 1,000,000 samples | 2.834 s | 235.21 dB | 235.61 dB | 0.051 / 37.163 ms | 0.051 / 14.470 ms |
| Mixed Phase 4M | 4,000,001 | 2,000,000 samples | 5.669 s | 253.25 dB | 254.42 dB | 0.051 / 37.299 ms | 0.051 / 14.493 ms |
Measured at x2, x4, x8, x16 and x32 for all seven depths. Every Mixed Phase depth rejects images by at least 120 dB at every rate. The worst results: Mixed Phase 6K 124.286099 dB at native x8 (124.286 dB), Mixed Phase Extended 500K 196.254044 dB at x32 (196.25 dB). At 1M ×8: 217.16 dB dense-stopband rejection, 217.25 dB at the 21→23.1 kHz image, 0.000000857 dB largest passband error and 0.181 ms group-delay spread; the impulse stays above −60 dB for 0.051 ms before and 37.523 ms after its peak, the step for 0.051 ms before and 14.195 ms after; 20 ms before the peak it sits at −105.19 dB.
Desk and pocket
The same filters run on Windows, Mac, iPhone and iPad. Each one plays to a DAC you plug in, or streams over Wi-Fi to the UPnP network streamers you already own.
Windows and Mac: up to 4M taps live and 16M for files, to your USB DAC through ASIO, WASAPI or Core Audio, or to Diretta on Windows.
iPhone up to 1M taps, M-series iPad up to 4M, out to a USB-C DAC. Free below 500,000 taps.
Stream over Wi-Fi to any UPnP network streamer, from the computer or the iPhone. Or let a UPnP app such as mconnect send its music through HeadMania, and steer the computer from your phone.
App Store screens. Swipe or scroll sideways.
Start the server and HeadMania becomes a music server for your home: any browser on your network opens the full app at your computer's address, port 8080. The iPhone and iPad app finds your computer by itself and controls it, and so does HeadMania.org Remote for Android, now in Google Play testing. The phone chooses; the computer plays.
Android remote: to join the test, ask in the Telegram group, or join the testers group and open the Google Play test on your phone.
This app can also control HeadMania on your Windows or Mac computer: tap the remote button, pick a computer, and its library, filters and settings appear here. It finds computers on your network by itself, or you add one by address. The computer plays; this device chooses.
The same engine on iPhone and iPad: 1 million taps on iPhone and 4 million on iPad Pro M5 and other M-series iPads, live, to a USB-C DAC or a network streamer. It also measures your room with the iPhone's microphone and remote-controls HeadMania on your computer. Get it on the App Store.
Your music
Your own files and CUE albums, HIGHRESAUDIO with your subscription, lossless radio checked station by station, or the sound of any other app. A UPnP app such as mconnect can send its music through HeadMania too.
Add folders and files; albums, CUE sheets and a library view that opens straight to your albums.
All platformsAdd a network share as a library folder (on Windows, \\server\share). Network tracks play, seek and stop like local files; a sleeping NAS shows "Waking up…" instead of a frozen app, and Windows' own sign-in window keeps your password away from HeadMania.
Windows & MacStations checked on the wire: FLAC only, CD quality or better, over HTTPS. Each one shows its format and data use.
All platformsCapture what another player, a browser or a game plays and send it through the filters. With Organik Live the filter itself adds under 0.2 ms; the capture's own buffers add tens of milliseconds.
Windows & MacHeadMania is a network renderer too: mconnect, JPLAY and other control apps send their music through it.
All platformsThe hi-res catalogue with your own subscription.
More about the partnershipAll platforms
Your folders, playlists and CUE-sheet albums, as tracks. The app window and every browser on your network see the same library, and folders can be added from a phone.
Every station is checked on the wire: FLAC, CD quality or better, HTTPS only. Each shows its data use per hour.
On Windows and macOS, HeadMania can capture what other apps play, such as a streaming service or a browser, and filter it on the way to your DAC.
JPLAY, Audirvana, mconnect and other UPnP apps can send music straight to HeadMania as a network renderer.
Play files in place from the device, iCloud Drive or Files, or copy them over with AirDrop or USB: FLAC, WAV, ALAC, AIFF, MP3, and CUE albums.
Every station is checked on the wire: FLAC over HTTPS. A switch decides whether radio may use mobile data.
JPLAY, mconnect and other control points send music to HeadMania on this device.
Trade-off: captured audio has already passed through the system mixer, so capture improves the output path only. UPnP apps keep the mixer out entirely.
Out to your DAC
A USB DAC on Windows and Mac, Diretta on Windows, a USB-C DAC on iPhone and iPad, or any UPnP streamer over Wi-Fi. PCM from ×2 to ×32 on Windows and Mac, and up to ×16 on iPhone and iPad: 768 kHz from a 48 kHz source. DSD from HeadMania's own modulator up to DSD512, DSD1024 on Windows with native ASIO, and DSD2048 as an experiment on Windows. iPhone and iPad send DSD to a UPnP streamer that accepts it; over USB-C they send PCM.
USB DACs on Windows (WASAPI exclusive or ASIO) and macOS (Core Audio).
Any UPnP renderer, in PCM, native DSD, or DoP for a renderer you mark as DoP-capable.
On Windows, a Diretta target as a third output. Needs the Diretta Host plugin.
Plays PCM to a connected USB-C DAC. DSD over USB is not yet available on iPhone and iPad.
Relays the filtered stream to a UPnP renderer. Native DSD follows what the renderer supports; DoP only for a renderer you mark as DoP-capable.
Choose ×2 to ×32 (×16 on iPhone and iPad). Your source keeps its rate family: 96 kHz ×8 plays at 384 kHz, 88.2 kHz ×8 at 352.8 kHz, and the Play tab reports any lower rate the DAC settles on.
HeadMania turns the upsampled signal into DSD with its own modulator. Send it native, or as DoP for a DAC that needs DSD wrapped in PCM; a DAC that does not understand DoP plays it as full-scale noise. An optional EC modulator (9th-order, extended compensation; off by default) lowers the noise further from DSD64 to DSD512, at up to about 1.6× the modulator's processor work. DSD1024 and DSD2048 run on Windows over native ASIO, with the standard modulator.
| Rate | Standard | EC modulator | Status |
|---|---|---|---|
| DSD64 | 111 dB | 122 dB | Regular |
| DSD128 | 154 dB | 174 dB | Regular |
| DSD256 | 172 dB | 228 dB | Regular |
| DSD512 | 194 dB | ≥ 240 dB | Native only |
| DSD1024 | 224 dB | not offered (processor-bound) | Windows, native ASIO |
| DSD2048 | > 240 dB | not offered (processor-bound) | Windows, native ASIO, AVX2 processor |
In-band dynamic range of the modulator output, 1 kHz tone, 20 Hz–20 kHz. What reaches the DAC is capped near 157 dB by the interpolator, and a DAC's analog stage typically sits near 120 dB: read the ladder as headroom, not as an audible difference.
| Your DAC | Start with |
|---|---|
| True 1-bit or hybrid multibit | DSD256 |
| R2R or NOS (e.g. Rockna, Denafrips) | PCM |
| ESS or AKM chip | PCM; the chip converts DSD internally |
| FPGA (e.g. Chord) | PCM and the DAC's own filter |
Headphones and sound shaping
AutoEQ profiles from the open AutoEq database, your own parametric EQ, crossfeed and effects, all loaded with each headphone. Level protection guards against peaks that form between samples; DAC profiles give starting settings for your DAC type.
AutoEQ corrects your headphone model with profiles from AutoEq, Jaakko Pasanen's open-source project built on measurements by independent labs; each profile names its lab. Compare measurements, fine-tune bands by number, and filter by over-ear or in-ear. Corrections are level-neutral.
Parametric EQ gives you ten bands to set yourself. Paste Equalizer APO or oratory1990 settings, save profiles, export and import them.
Bass Thump, Vocals Presence, Speed, Harmonic Exciter, Parallel Detail, Crossfeed and Soundstage Width. With every slider at zero they are bypassed and the signal is untouched.
Optional. 10 ms lookahead, −0.13 dBFS ceiling; true-peak qualified from ×8 (about 0.17 dB residual at ×8). On loud masters it trims dynamic range slightly.
Pick your DAC type and HeadMania matches the level margin and dither: rounding for R2R at 24-bit, noise-shaped dither for delta-sigma, flat TPDF for 1-bit, none for 32-bit float.
Action: Bass 0.60, Speed 0.20, Harmonic 0.10. Dialogue: Vocals 0.75, Detail 0.20, Bass 0.10. Night: Detail 0.80, Vocals 0.20.
No black box
The Play tab shows the filter, rate and engine actually running, any step-down and why, and the reason a stream stopped.
The Play tab shows the filter, rate and engine actually running, any step-down and why, peaks above full scale on the waveform, and the reason a stream stopped. The performance panel shows real-time headroom with processor, graphics and memory use, and the Debug tab copies or saves a log for support.
Neuromancer · Room correction
Two stages that go beyond the upsampling filter. Neuromancer, a neural model with 8.5 million learned values, reshapes the music as it plays, before the filter, on your graphics card or the Neural Engine. Room correction simulates your speakers and seat in 3-D from the floor plan alone, then cuts your room's bass peaks with a measurement from an iPhone or a calibrated microphone. Both are optional and clearly marked: Neuromancer is experimental, room correction is in testing.
Reshapes the sound as the song plays: about 730 billion operations for every second of music, on your graphics card or the Neural Engine. The app tests your device first, and Original, which plays the track untouched, stays one tap away.
Neuromancer is an optional neural stage, a model with 8.5 million learned values, that reshapes the sound as the song plays, before the upsampling filter. Choose Neuromancer, Awakened or Revelio, each with a lighter Eco version. With Revelio, Revelio's Pulse lifts drum attacks above 2.5 kHz and Unleashed shapes them in three bands. It changes the signal on purpose; Original plays the track untouched.
At full size it runs about 730 billion operations for every second of music and re-analyses every frequency 172 times a second. Measured on our test devices, an NVIDIA RTX 5090 processes a quarter-second of music in 14.5 ms (17× faster than real time) and an iPhone 18 Pro Max in up to 93 ms (2.7×). The Eco models need much less.
Runs on a Windows graphics card (it tests the card first; on NVIDIA cards an optional pack of about 1 GB speeds it up) or an Apple-silicon Mac with macOS 27. Neural Master is a lighter option that sets a few controls once per track.
Free, and not part of Signature Edition. Runs on the Neural Engine and processor, with iOS or iPadOS 27.
Trade-off: tracks start about 2.75 s late (about 0.3 s more above 44.1 kHz). Until a test shows the device keeps up in real time, tracks play untreated. It accepts sources from 44.1 to 192 kHz, and 352.8 kHz.
Every Neuromancer model was trained on recordings from:
Neuromancer Awakened and Neuromancer Revelio were also trained on stereo masters from:
To the extent a trained model is an adaptation of that material, these model files are offered under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International licence: https://creativecommons.org/licenses/by-nc-sa/4.0/. They are distributed free of charge and must not be sold or placed behind a paywall.
From your room's shape, speakers and seat, HeadMania simulates the sound in 3-D and ranks where the speakers and the seat could go. Measure with an iPhone or a calibrated microphone, and it cuts the room's bass peaks.
Two independent routes. Simulate: from your room's shape, size, speaker positions and seat, HeadMania simulates the sound travelling from both speakers to your listening spot in 3-D, at ear height. It ranks where the speakers and the seat could go, and estimates a correction for where you sit from the room plan alone, with no microphone. Measure: record your speakers at the seat with the iPhone's microphone, or an external measurement microphone and its calibration file. Sweeps play through a local DAC or a UPnP streamer on a computer, and through a UPnP streamer on iPhone and iPad; HeadMania allows for the network's delay. Either way, the correction cuts the room's bass peaks. Rooms can be saved and moved between devices.
Status: the Room tab is still in testing. Ringing reduction is a preview: calculated, not yet applied.
Feature index
Run HeadMania on a PC or Mac as your network audio processor; browse and play your library and HIGHRESAUDIO from a browser or the iPhone app.
Windows & MacWhole albums and lists play on from the computer while your phone sleeps.
Windows & MacNative DSD where the player accepts it, or DoP for a player you mark as DoP-capable, from the computer, the iPhone or the iPad.
All platformsA Diretta target as a third output, next to your DAC and UPnP players.
WindowsSteer the computer from the iPhone app or any browser on your network; an Android remote is in testing.
All platformsCover art with Previous and Next; albums keep playing while the phone is locked.
iPhone & iPadOpen AutoEq profiles with pre-amp and compare; your own bands with paste, copy, save, export and import.
All platformsA guard for peaks that form between samples, plus an optional output limiter (−0.13 dBFS ceiling), against clipping from loud masters.
All platformsPick your DAC type and HeadMania matches the level margin and dither: rounding for R2R, noise-shaped dither for delta-sigma, flat dither for 1-bit.
All platformsSimulate your room, find the best seat and speaker layout, measure with an iPhone or a calibrated microphone, reduce ringing.
In testingReal-time headroom for the filter that is running, so you know what your computer can carry.
Windows & MacDesktop updates are signed and install only when you choose.
Windows & MacPrice
Free
Every filter, every output and HIGHRESAUDIO with your own subscription.
Free
Every filter below 500K taps, Neuromancer, room measuring and the remote.
One payment
Adds the larger filters, up to 1M on iPhone and 4M on M-series iPads, and applies your room correction.
One payment, 30-day trial; the App Store shows the price for your country.
On Windows and Mac, unzip the download and open the app inside; after that, updates arrive in the app, signed and installed when you choose. All versions and release notes: GitHub releases.
Free: every filter below 500K taps, Neuromancer, room measuring and the remote. Signature Edition adds the larger filters, up to 4M on M-series iPads, and applies your room correction. One payment, with a 30-day trial; purchases can be restored and codes redeemed.
Spec sheet
| Item | Value | Platform |
|---|---|---|
| Engine | 64-bit floating point | All platforms |
| Reference Precision | Double-double (about 106-bit) at 4M, 8M and 16M taps, for files and the server | Windows & Mac |
| Live filters | Up to 4M taps; 1M on iPhone, 4M on M-series iPads | All platforms |
| Filter families | Linear, Organik, Mixed Phase, Organik Touch & Pulse, Reference Precision, Organik Live | All; the last two Windows & Mac |
| Low latency | Organik Live 0.13 to 0.18 ms; block filters 3.7 to 9.8 ms from a 44.1 kHz source | Windows & Mac |
| PCM out | ×2 to ×32; ×16 on iPhone and iPad (768 kHz from 48 kHz) | All platforms |
| DSD out | DSD64 to DSD512; DSD1024 over native ASIO; DSD2048 experimental (AVX2 processor) | Windows & Mac; iPhone & iPad to UPnP streamers; DSD1024/2048 Windows |
| USB DAC | WASAPI exclusive or ASIO; Core Audio | Windows; Mac |
| USB-C DAC | PCM to a connected USB-C DAC | iPhone & iPad |
| Network streamers | Any UPnP renderer, over Wi-Fi or cable: PCM, native DSD, or DoP | All platforms |
| Diretta | A Diretta target as a third output (Diretta Host plugin) | Windows |
| Sources | Library and CUE albums, NAS or shared folder, lossless radio, other apps' sound, UPnP apps, HIGHRESAUDIO | All; NAS and capture Windows & Mac |
| Headphones | AutoEq profiles, parametric EQ, crossfeed and effects per headphone | All platforms |
| Output limiter | Optional; 10 ms lookahead, −0.13 dBFS ceiling | All platforms |
| Neuromancer | Neural stage with 8.5 million learned values (experimental) | All platforms |
| Room correction | Simulate, measure, cut bass peaks (in testing) | All platforms |
Questions
On Windows and Mac, yes, with every filter and output. On iPhone and iPad, every filter below 500,000 taps is free, with Neuromancer, room measuring and the remote. The one-time Signature Edition adds the larger filters and applies your room correction.
A USB DAC (USB-C on iPhone and iPad), or nothing at all if you stream over Wi-Fi to a UPnP network streamer. Headphones go on your DAC. On Windows, Diretta is a third option. On iPhone and iPad, Bluetooth headphones work too, without the upsampling.
HeadMania sends to UPnP renderers in PCM at the rates your streamer accepts, native DSD where it accepts DSD, or DoP for a streamer you mark as DoP-capable.
Yes, to a UPnP network streamer: native DSD where the streamer accepts it, or DoP for one you mark as DoP-capable. Over USB-C, iPhone and iPad send PCM, up to ×16 (768 kHz from a 48 kHz source); DSD over USB is not yet available there.
On Windows, a graphics card: HeadMania tests it first, and on NVIDIA cards an optional pack of about 1 GB speeds it up. On a Mac, Apple silicon with macOS 27; on iPhone and iPad, iOS or iPadOS 27, on the Neural Engine. Tracks start about 2.75 s later, and until the test shows your device keeps up, they play untreated. It is experimental, optional and free, not part of Signature Edition.
Yes, with the Low latency filters. Organik Live adds 0.13 to 0.18 ms; the capture of another app's sound adds tens of milliseconds of its own.
Only for HIGHRESAUDIO streaming. Your own music, lossless radio, other apps and UPnP apps work without one.
HIGHRESAUDIO tracks are held in memory while they play; no music is written to disk. This page sets no cookies and loads nothing from other sites. The app's privacy policy has the details.
An Android remote for the computer is in testing on Google Play. The upsampler itself runs on Windows, Mac, iPhone and iPad.
On Windows and Mac, updates are signed and install only when you choose. On iPhone and iPad, the App Store updates the app.
Not sure? Start with Organik 2M. Pick Linear for a DAC with no filter of its own (NOS, R2R).
In the press
Who makes it
HeadMania.org Upsampler is independent software by Dan Gheorghe, who reviews DACs, headphones and speakers at headmania.org. HIGHRESAUDIO streaming is built in with HIGHRESAUDIO, Berlin. Questions go to the Telegram group, bugs to GitHub, and Patreon keeps it going.
On Windows and Mac, unzip the download and open the app inside; after that, updates arrive in the app, signed and installed when you choose. All versions and release notes: GitHub releases.