Part One: Why I Wanted to Build Mossca
My Listening Habits
I have tried online streaming services before.
In China, the two biggest ones are QQ Music and NetEase Cloud Music. To be honest, I do not use 90% of their features. I just want to listen to music quietly. I do not need social features, comments, playlists pushed everywhere, ads, or a bunch of unrelated things that have nothing to do with listening. I got tired of those bloated "all-in-one" products. There is also the copyright problem: music I like may suddenly disappear one day.
Spotify is relatively closer to what I want. I still use it today, especially for checking out new albums and niche releases. It is convenient and simple enough, although the sound quality is not ideal. For me, Spotify is more like a listening station. I use it to discover new music, and if I really like something, I will buy the album and keep it in my own collection. Spotify's UI is also the closest to my taste among streaming apps.
Apple Music is another option. I tried it for a while. At first, the music coverage was not enough for what I wanted. Later it improved, but for casual listening I still found Spotify more convenient, especially because I can just open Spotify Web and start listening. Apple Music's app experience and UI are fine, but not especially exciting. It does provide lossless audio, but I eventually stopped using it as my main option because it still feels like a product built for the mass streaming audience. For me, Spotify is already enough for that role.
So my main listening space is still local music.
It feels a bit like listening to a Walkman or CD player when I was younger. Quiet, personal, simple, and pure. It is my own music, and nobody interrupts me. I also know quite clearly what I want to listen to. For the music I truly love, I definitely want to own and keep it permanently.
Right now, my Samsung T7 2TB temporary drive is already full. I also have more than 4TB of lossless albums on Baidu Cloud and Quark Cloud, and the library is still growing. I am also planning to build a NAS system soon. My music comes from many places and includes many formats, mostly DSD, WAV, and FLAC. I listen to a lot of rock, pop, and classical music, from the 1990s to modern releases, from Europe and America to Japan.
The Search for a Local Music Player
After I started listening to lossless local music, I began looking for local lossless music players. My needs were actually very simple:
- Good sound quality.
- A clean and tasteful UI, not outdated.
- Friendly to large music libraries, because I am not only going to listen to a few albums.
- Simple, without too many complicated or irrelevant features.
Audirvāna
The first one I tried was Audirvāna. Anyone who listens to local music will probably run into it sooner or later.
Its UI is acceptable, and the feature set is moderate. I would not call it bloated, but I would not call it simple either. Many settings still feel a bit engineering-oriented, and for me, most of them are not useful.
Its support for large album libraries is decent. After importing, scrolling and scanning are reasonably smooth. But some albums were parsed incorrectly and showed garbled characters like "???". I also do not really like its default album list. To me, album covers are important, almost like vinyl covers. They should not just be tiny functional thumbnails in a list.
The most important part is sound quality. Although Audirvāna is known for this, in my own listening tests it felt a little dull and dry. This is just my personal impression, not a universal standard. The sound simply did not leave a strong impression on me.
Roon
Then I tried Roon. Its white theme is modern and relatively clean. But after using it, I felt that it emphasizes the home media center scenario more than sound quality itself. The main interface has quite a lot of features. The playback side is fine, but not that different in daily use. In my own music library, its metadata recognition was actually worse than Audirvāna, and the list view experience was also not ideal. It just does not feel like a product built around my kind of local playback scenario.
Its album information and editorial content are quite nice. But for where I am now, I still do not fully feel its value.
Swinsian
Swinsian feels a lot like iTunes. It is very easy to use. It does not really focus on sound quality either, and feels more like a modern version of iTunes. But it is fast. Playback, browsing, and large library navigation all feel very responsive, and that impressed me. Although it does not advertise itself as an audiophile player, the sound is actually quite pleasant: warm, layered, and easy to listen to. During development, if I need background music, I often leave Swinsian playing.
There are two things I do not love. First, the main interface layout is too complex for me, with too many window areas and panels. My core use case is still listening to music, not managing a cockpit. Some parts are also not easy to control or hide. This is probably related to its product positioning. After all, metadata editing is not my core daily use case.
Second, it still has many features, and honestly most of them are not useful to me. When playing through an external DAC, I also did not find DoP support. For me, that is a problem, because I do listen to DSD.
HQPlayer
Then there is HQPlayer. I saw it mentioned often in audiophile forums. It seems very hardcore, with deep experience in audio processing. I installed it and tried it once, then it basically just stayed on my computer. It may have many high-quality algorithms, but the interface feels too old. It feels more like an old precision instrument than a modern music app. The UI and playback structure already made me give up before I could seriously test the sound quality. I simply want to listen to music. I do not want to spend my time tweaking.
foobar2000
foobar2000 is another classic player. I used it on Windows many years ago, so I also tried the Mac version. But it is clearly not as convenient as the Windows version. It has many plugins and extensions, but that also means I need to spend time researching and configuring it. Like HQPlayer, it feels more like an engineer's product.
Apple Music as a Local Library Player
Yes, I tried it because I generally trust Apple's UI and experience. But when I imported my library of more than 3,000 albums, the list already started to feel noticeably slow. Some details also felt strange for no obvious reason. My feeling was still the same: Apple Music is not primarily about sound quality, and not primarily about local music. Streaming is its real core.
So I Decided to Build My Own
After trying all of these, and after giving up on some other players even during installation or early testing, I realized that very few products really matched what I wanted. So I thought: fine, I will build my own.
Based on my own needs, I summarized the key points:
- Sound quality matters. There are many standards for what "good sound" means. Personally, I am more interested in hearing the music itself as it is. There are already many music apps and even hardware products that add processing, coloration, or enhancement. I prefer something closer to a monitor: reduce unnecessary smart processing, avoid changing the music for no reason, and present it as faithfully as possible.
- UI matters. A music player is not only about sound. The interface also matters, especially for an app you may use every day. Visual feeling is part of the listening experience. Album covers matter, just like vinyl covers matter.
- Simplicity matters. Everything should serve listening itself. I do not want every possible feature to be added, exposed, configured, and turned into a control panel.
- Large music library support matters. A good local player should not only work well with a few albums.
- Scenario support also matters. Most of the time, I work in front of my computer. So the desktop experience is my first priority. I also have a DAC and monitor headphones on my desk, so external DAC support is important, whether I am using headphones or speakers. If I build a NAS later, support for a home music library will become even more important.
Other details also matter, such as DSD support and lyrics support.
So I started building Mossca.
Part Two: How a Good Local Music Player Works on Mac
Next, I want to explain, in my own understanding, what it takes to build a good and complete local music player on Mac and iOS. I will try to keep it understandable. There is not much mysticism here. Once you understand the basic signal chain, many marketing terms become much easier to judge. You will know what really matters and what does not.
Let's Start with Mac
From the user's point of view, playing a song means:
- Import local music into the app library. This part is usually not difficult. The real questions are accuracy and coverage. Can the app correctly parse artist, album, cover art, track information, and cue-split tracks? After importing, can it browse smoothly, especially when there are many albums?
- Click play. For the user, this is the core moment. After that, the main question becomes: does it sound good? It looks simple, but the engineering behind it is quite complex.
From an engineering point of view, the Mac playback chain is roughly:
Decoding → App audio engine → macOS Core Audio / HAL → Output device
A simple factory analogy:
- Decoding means unpacking the raw material and restoring it into data that can be processed.
- The app audio engine decides whether to process it, how to queue it, and how to send it out.
- Core Audio / HAL is the system-level conveyor belt and scheduling layer. It sends the audio stream to the target device. In shared mode, it may also participate in mixing and sample-rate conversion.
- The output device or DAC receives the digital signal and turns it into analog sound.
Decoding
Decoding converts compressed or packaged audio streams into data the player can send into its audio engine. Unless a player builds its own decoder, most apps rely on mature existing solutions. As long as the decoding path is correct, this part is usually similar across players, unless an app intentionally performs resampling or other processing.
For lossless music, most formats eventually become PCM:
- FLAC → PCM
- ALAC → PCM
- APE → PCM
- WAV / AIFF → usually already PCM
- DSF / DFF → DSD bitstream, or converted to PCM
For Mac itself, DSD is not normally handled like PCM in the default system audio chain. A player usually either converts DSD to PCM, or uses DoP to package DSD inside a PCM-like carrier and send it to a DAC that supports DoP.
App Audio Engine
Before audio reaches the system layer, the player may process it. The app engine sends decoded audio data into its own playback pipeline. It manages play, pause, seek, track switching, buffering, volume, and playback state. It is the brain of the app's audio system.
This is also where many apps apply sound processing. Different apps make very different choices here. Common processing includes:
- EQ
- Crossfeed
- SRC, including high-quality sample-rate conversion algorithms
- DSP
In simple terms, this is the most common place where sound gets "cooked". Whether that cooking is good or bad depends on the chef. Some players add more seasoning. Some try to preserve the original taste of the ingredients. This is a product choice.
macOS Core Audio
Once the audio reaches this layer, it is in the system's territory. It must go through this layer before reaching the device.
This is where people often talk about exclusive mode and shared mode. In simple terms, exclusive mode means the app tries to take dedicated control of the audio device. The advantage is reducing interference from system-level mixing or resampling. Shared mode means the app shares the audio device with other apps.
When exclusive-style output is active, other apps may temporarily lose access to that audio device, or their audio may be routed elsewhere depending on the system and device behavior. On pure Mac playback, the audible difference may not always be dramatic. With an external DAC, the difference may be more obvious.
macOS does not have a single user-facing "Exclusive Mode" name like Windows WASAPI Exclusive. Players usually implement a similar experience through Hog Mode, device ownership, Audio HAL device binding, sample-rate switching, and related techniques. This is one of the differences between the two platforms.
Output
This is the easiest part to understand: headphones, speakers, and often a DAC in between. Different devices have very different capabilities, and adapting to DACs can be difficult.
Strictly speaking, a DAC is not just an "output layer". Its role depends on what the player sends to it. If the player sends PCM, the DAC converts digital PCM into analog sound. If the player sends DSD through DoP, the DAC recognizes the DoP marker, extracts the DSD bitstream, decodes DSD through its own path, and then converts it to analog.
DoP means DSD data is packaged inside a PCM-like carrier for transmission. A DAC that supports DoP recognizes the marker and extracts the original DSD bitstream. In theory, this is much closer to preserving DSD playback than converting DSD to PCM in the player.
Two Key Questions in the Audio Signal Chain
1. What Affects Sound Quality?
Anything that changes the music data itself can affect sound quality: upsampling, downsampling, SRC, digital volume, EQ, DSP, and the choice between shared and exclusive output.
From this point of view, the cleanest path is usually the one with the least unnecessary interference. But again, it is like cooking. Some people like more seasoning; some people like the original taste. Except for clearly negative processing, different players will make different trade-offs.
SRC, digital volume, EQ, and DSP directly change digital audio data. Exclusive/shared mode is more about whether the audio goes through system mixing, whether the system resamples it, and whether the output sample rate can match the device and source.
2. Bit-Perfect and Sample-Rate Handling
In the signal chain, common input parameters include bit depth and sample rate, such as 24-bit / 96 kHz. Because of hardware limits and processing choices, the final output bit depth and sample rate may be different.
Two details are especially important:
- Try to keep the output data as close as possible to the decoded source data. One important part of this is matching the sample rate.
- If conversion is necessary, same-family conversion is usually preferable: for example, 44.1 → 88.2 instead of 44.1 → 96.
Higher Sample Rate Does Not Automatically Mean Better Sound
Sample rate means how many samples digital audio records per second. A higher sample rate does not automatically mean "better sound". It mainly affects:
- The highest representable frequency
- Digital filter design space
- Headroom for post-processing
- Whether the system or device needs to resample
According to the Nyquist theorem, 44.1 kHz can theoretically cover up to around 22.05 kHz, while human hearing usually tops out around 20 kHz. So CD's 44.1 kHz was not chosen randomly. It was designed to cover the human hearing range.
The value of higher sample rates is more often in: more processing room during recording and production, more relaxed filter design, better behavior on some DACs or playback chains, and reducing side effects in some processing stages. But that does not mean every 192 kHz file sounds better than every 44.1 kHz file.
Of course, psychologically, many of us still prefer higher numbers when everything else is equal. I am the same. It is a bit like photography: 20+ megapixels may already be enough, but if there are more pixels, I still feel more comfortable choosing the higher number.
Why Bit-Perfect Matters
Bit-perfect means: the digital audio data sent by the player to the device should stay as close as possible to the original decoded PCM data from the source file.
More specifically:
- No system resampling
- No digital volume change
- No EQ
- No ReplayGain
- No DSP
- No forced format conversion
- No system mixing
- Sample rate matching the source when possible
- No unnecessary changes to channel layout or bit depth
Bit-perfect does not matter because it always sounds best. It matters because it provides the cleanest and most verifiable baseline. In simple terms: whatever the music originally is, I try to present it that way.
Its value has three parts:
First, it reduces uncertainty. If a 44.1 kHz FLAC file is forced by the system into 48 kHz, then mixed with other system audio, then scaled by digital volume, what you hear is no longer just the file itself. It is the file plus system processing, player processing, and device behavior. That may not sound bad. But it is not transparent.
Second, it avoids low-quality processing. Resampling, digital volume, and DSP are not evil. The question is: who is doing it, how is it being done, how good is the implementation, and is it necessary? Some system SRC implementations are good enough in many cases. But if a player can avoid unnecessary SRC, it avoids a possible source of distortion.
Third, it builds trust. For a local music player, bit-perfect is a professional baseline. Do not secretly change the audio. If processing happens, tell the user. If the player cannot output the source as-is, clearly show fallback, resampled, or degraded status.
Bit-perfect does not always mean more pleasant sound. Monitor headphones or speakers are not always what people find most flattering either. Some people find them too plain. Again, it is a product choice.
Why Same-Family Conversion Matters
Same-family conversion means:
- 44.1 kHz source files should preferably convert to 88.2 / 176.4 / 352.8
- 48 kHz source files should preferably convert to 96 / 192 / 384
The reason is mathematical simplicity. 44.1 → 88.2 is 2x. 44.1 → 176.4 is 4x. 48 → 96 is 2x. 48 → 192 is 4x. These integer-multiple conversions are relatively direct.
But 44.1 → 48 is a non-integer conversion. The ratio is 160:147. That means the resampler needs more complex interpolation, filtering, and phase handling to turn 44,100 samples into 48,000 samples.
This does not mean 44.1 → 48 must sound bad. A high-quality SRC can do it very well. But in principle: integer-multiple conversion is simpler; cross-family conversion is more complex; more complexity means more dependence on SRC quality; if it is unnecessary, avoid it.
So the principle is: 44.1 should preferably go to 88.2 or 176.4, not be forced into 48 or 96 without reason.
About Upsampling
Upsampling means converting lower sample-rate audio into a higher sample rate. For example: 44.1 → 88.2, 44.1 → 176.4, 48 → 96, 96 → 192.
Many people misunderstand this: bigger number = more information = better sound. That is wrong. Upsampling cannot create information that does not exist in the original file. A 44.1 kHz recording does not contain real 30 kHz or 40 kHz recorded content. Converting it to 192 kHz does not magically generate true ultrasonic information.
What upsampling really does is: insert new samples between existing samples, use filters to estimate the intermediate waveform, convert the data into a higher sample-rate format, and sometimes move digital filtering work from the DAC to the player. It may help, but not because it creates new information.
Possible benefits include: the player's SRC algorithm may be better than the DAC's internal upsampling, some DACs behave better with high sample-rate input, same-family integer upsampling makes later processing easier, DSP or EQ may have more working room at higher sample rates, and it may avoid low-quality system SRC.
But it may also bring no benefit, or even sound worse. Many fake lossless files are simply upsampled. The actual signal is limited, and the rest is basically guessed.
Now, iOS
The iOS platform is very different from Mac. There are two big differences.
First, mobile devices have more limits: battery, performance, thermal behavior, and system restrictions. They are not like desktop machines.
Second, the playback chain is different. iOS cannot use the same kind of exclusive output path as Mac. On iOS, the key system audio layer is AVAudioSession. On Mac, the core is Core Audio / HAL / output device handling. This is one of the biggest differences between iOS and Mac.
There is also a more important practical difference: user scenario. Mac feels more like a serious music playback terminal. It has enough performance, can integrate with home audio equipment, and can connect to external DACs. It is better suited for deeper, more focused listening.
iOS is naturally a mobile scenario. Users care more about battery, heat, lightness, and a simple mobile UI. They do not necessarily need extreme audio engineering at all costs. In many mobile scenarios, the output device is Bluetooth headphones, and the sound quality is already limited by the chain. Outdoor noise is also often high. Even with an external dongle DAC, power and size limits mean it cannot really compare to a desktop DAC with independent power.
So recently I have also been thinking about the real value of Mossca Mobile. While still trying to preserve good sound quality, we are considering a new direction. We will need to validate it after the first version is ready.
Some Lessons from Building Mossca So Far
1. Stable Rhythm and Stable Experience Matter
In the early stage, there were many basic features to build, so development moved very aggressively. If you look at the release notes on the website, there was basically a new version every week or two. Fast iteration helped, but it also left some architectural debt. In the last two versions, I have slowed the rhythm down and focused more on core stability and raising the baseline.
2. Focus on the Truly Core Experience
Earlier, to make the app feel more professional, I added some features that were not really core, such as a waveform progress bar. I have removed it in recent versions. These things add extra code complexity, and if they do not truly serve listening, they should go.
We also built Radar, a system for evaluating music library quality. It is valuable, but it is not part of the core playback path. So for now, it has been downgraded. The value of helping users evaluate library quality and identify fake lossless files will remain, but it should not interfere with the core listening experience.
3. Some Early Architecture Decisions Were Not Good Enough
The main reason is that I did not fully understand all playback scenarios early on. Some initial architecture choices were not ideal. I am now fixing them step by step. For example, some earlier features used fallback paths because of performance concerns, even if the experience was not ideal. These debts are now being paid back one by one.
Where Mossca Is Now
Mossca now covers the core path from importing and parsing a music library to playback, across macOS, Windows, and iPhone / iPad.
Music libraries and connections: In addition to local libraries, Mossca can connect to music stored over SMB and WebDAV, as well as Mossca Server, Navidrome / Subsonic, Jellyfin, Plex, and Emby. The format coverage includes the major lossless formats, including DSD.
Playback: On macOS, Mossca supports compatible external DAC output, exclusive output, DSD through DoP, and high-quality sample-rate conversion. On iPhone and iPad, playback follows the capabilities and routing of iOS; the relevant boundaries around exclusive access and Bit-Perfect output are explained separately in Listening Notes. Signal-path information and online lyrics matching are also available.
Scenario support: Mossca for Windows is now available alongside the Mac and iPhone / iPad apps. Mossca Server is available for free deployment on a NAS or home server, providing a web player and a music-library connection for Mossca clients.
Library management: Metadata editing is available for supported files, so maintaining tags can remain part of the listening workflow rather than a separate chore.
At the same time, core playback stability, response speed, and UI details, especially after adding multiple themes, will continue to be refined. The current state is still not ideal enough.
If you have any feedback or thoughts, feel free to email me. I welcome criticism and suggestions. I am also learning and improving quickly.