Tool for musicians

[goermezer]

Love this idea. Actually a very usable and helpful tool. I`m really happy that I found it. In case you are looking for further development ideas, here are mine from a musicians point of view:

• A control in the player to slow down the playback would be extremely helpful. Just mute the guitar track and practice with stemdeck and scores on paper for example.
• Spotify support - of course with same legal restrictions like youtube
• Config to save the jobs permanently - perhaps something like STEMDECK_JOB_TTL_SECONDS=0
• Mobile view does not render well on a phone. But this use case is more or less obsolete because of streaming wav files which are too big for mobile connections.

Technical:

• Currently the browser streams the wav files to the browser. For streaming a lightweight format like flac, mp3 or similar could reduce the the necessary bandwidth for streaming.
• More GPU support like Intel, AMD would be cool. But I think I should stop here :-)

[thcp]

Thanks for the thoughtful list, @goermezer — really appreciated 🙏

Job persistence ("save the jobs permanently"): this is now sorted.

• Desktop app: the 24-hour cleanup no longer touches your library at all — tracks stay until you delete them (via Trash). No config needed. Shipping in the next release (0.7.0 Alpha 18). (fix: don't auto-purge the desktop library (skip job TTL sweep) #229)
• Self-hosted / server: the cleanup stays on by default for disk hygiene, but you can set STEMDECK_JOB_TTL_SECONDS to a large value (it's in seconds) to keep jobs around for as long as you like. One gotcha: =0 won't disable it — the value is clamped to a 5-minute minimum — so use a big number instead.

Spotify support: we've looked into this, and we won't be adding it, due to the legal issues around it. There's no clean way to pull audio from Spotify — its streams are DRM-protected, and the only tools that actually rip them circumvent that DRM (a Terms-of-Service violation that also runs into anti-circumvention law like DMCA §1201). The popular "Spotify downloaders" (e.g. spotDL) don't really download from Spotify either: they read the track metadata and fetch a YouTube match instead — YouTube under the hood, with the added risk of grabbing the wrong/alternate version. Given StemDeck's local, privacy-first, not-a-downloader stance, we'd rather not take the project there. Pasting a YouTube or SoundCloud link remains the supported way to bring in source audio.

[thcp]

Ah, amazing, the video helped! basically memory issue, check #234.

next release will be coming up during lunch time.

[thcp]

@goermezer , 0.8.0 Alpha 3 is out: https://github.com/stemdeckapp/stemdeck/releases/tag/v0.8.0-alpha.3

This release includes all the mobile improvements we discussed, plus one bigger one that took a bit more work to get right.

The crash and the stuck loading screen are fixed. Loading a long track on mobile Safari no longer crashes the tab, and the mixer no longer gets stuck on "Preparing audio" after you reopen the app.

But the bigger change is how audio loads. Previously, the mobile player had to download the entire audio file for every stem before you could press play - on a 5-minute track with 4 stems that could be hundreds of MB sitting in memory at once, which is what caused the OOM crash on your device.

The new approach works the same way Spotify and most modern audio players handle this: instead of downloading everything upfront, the player fetches audio in small 10-second chunks and starts playing as soon as the first chunk arrives. While you listen to one chunk, the next one is already being fetched in the background. The result is that audio starts in a couple of seconds on WiFi, and memory usage stays around 28 MB regardless of track length - no more length limit either.

The play/pause button, seek, and prev/next navigation also got fixes in this release.

Let me know how it goes on your end.

[goermezer]

Many improvements. Beautiful!

I can confirm, crash and stuck loading screen does not happen anymore. Thanks to the new loading approach the mobile app has a much better experience now. It feels quite snappy. Audio loads in 1-2 seconds with wifi connection. With cellular it takes ~20 seconds for the first chunk (1 stem). And prev/next works well too.

[plan4uga]

Hi! I really like your app! Can you explain how can I switch to GPU compute, it greyed out for me. I have latest version installed. It's NVIDIA package. So everything done right I guess, but It still doesnt let me. I have 2070 Super. Maybe it's unsupported? Maybe clean install can help. How can I do it? I don't know) I've been folowing your app since release 0.5 or 6 and really great to see constant updates and improvements. Good luck on your way!

[thcp]

@plan4uga man, really sorry but I for buried by emails and in managed to check your message now. Did you tried the latest version?

[thcp]

Hey @goermezer, glad to hear Alpha 3 is working well! And good to know cellular is usable too, even if slower. Alpha 4 reduced the chunk size from 10s to 5s so the first audio should arrive a bit faster on slower connections now.

One more update to close the loop on your original list: playback speed control shipped in 0.8.0 Alpha 4.

You can now slow down or speed up a track with the speed slider in the player. It uses a pitch-preserving algorithm (WSOLA, the same technique SoundTouch uses), so voices and instruments stay at their original pitch regardless of tempo. Practicing guitar at 0.75x still sounds like the original recording, just slower. Works on both desktop and mobile.

On Intel and AMD GPU support: this one is genuinely complex to add properly. GPU-accelerated ML inference in Python is almost entirely built around the CUDA/PyTorch stack, which is NVIDIA-specific. The alternatives exist: ROCm for AMD and OpenVINO/IPEX for Intel, but they each require a separate build pipeline, platform-specific driver dependencies, and dedicated hardware in CI to test against. That means each GPU vendor is effectively its own artifact, its own compatibility matrix, and ongoing maintenance burden. We are keeping focus for now, but it is absolutely on the list for the future if we can get access to the right hardware. It is not forgotten.

So from everything you raised:

• Playback speed control: done, with pitch preservation (Alpha 4)
• Job persistence: done (desktop keeps tracks until you delete them, server has STEMDECK_JOB_TTL_SECONDS)
• Mobile view: done and improving across every alpha (1 through 4)
• Spotify support: not happening for legal reasons (explained above)
• Streaming format (FLAC/MP3): chunked WAV over Range requests gets the same low-bandwidth benefit without re-encoding; considered addressed
• Intel/AMD GPU: future roadmap item, technical challenges explained above

I think that genuinely covers everything you brought up. I really want to say thank you for the quality of your engagement here. Your feedback was specific and actionable, you tested every release, and when we needed more detail you recorded a screen capture. That kind of collaboration is rare and it made StemDeck concretely better. Big thank you for your ideas and for helping push the project forward.

If you want to follow along with what we are working on between releases, we share weekly updates on Instagram at @stemdeckapp. Always happy to see familiar faces there :)

[goermezer]

Wow, impressive! Never happened before that all wishes come true :-). I understand, that the GPU topic is challenging. I think most users can live with it. Most of the time it it doesn`t take too long to separate.

But I have two findings:

• The pitch preserving algorithm seems not to be working for me. Neither on mobile nor on desktop.
• As novice guitarist slowing down playback by 1%-20% in single percent increments is most of the time what I`m using. Sometimes practicing very fast songs even 30% or 40% happen. But not slower. Speeding up never happens. So my personal requirement would be:
  • Maximum speed: 1x
  • Minimum speed: 0.6x
  • Using slider set something in between in single % increments.
  • Perhaps a setting to set min, max and increment could be an idea. Not sure, but people have different preferences.

[thcp]

haha, no worries ! this is a little side project that I'm having tons of fun with it, either by the technical side or by interacting with people :)

I will be looking into those sugestions, i understand your usecase and it make sense for me 👍🏻

[thcp]

Coming back to this one because most of it landed.

Slower playback is in the transport, so you can drop the tempo and practice against a muted guitar. The library is permanent by default now, no TTL to configure. And there is a proper phone UI, which loads automatically when you open StemDeck on a phone.

Spotify is the one I am not going to do. There is no way to get the audio that respects their terms, so it would mean shipping something I could not stand behind.

Thanks for the original list. It aged well.