BubbleUPnP > HQPlayer OS repeats the beginning of tracks

Miska · January 11

13 hours ago, ohshitgorillas said:

This only happens with hi res material, not with redbook stuff. I'd assumed that hi res material was actually easier to process, given that there are more original samples so hqplayer has less interpolation to do. Am I wrong?

Depends on what you have selected as Nx filter. Any two-stage poly-sinc filter is heavier as source rate increases.

13 hours ago, ohshitgorillas said:

AVX2 build. Not sure about the Multicore DSP setting, as I can't find it in the HQPlayer Embedded web interface.

Not all settings in Embedded are in the web interface because Embedded is for hardware manufacturers building streamers. So the manufacturer knows their hardware and how optimally set up HQPlayer for their hardware. And end user purchasing such hardware who uses the web interface doesn't need to know about such details.

13 hours ago, ohshitgorillas said:

On that note, where is the HQPlayer Embedded manual located?

As help pages in the web interface. Help link is near top right corner of each corresponding web interface page.

13 hours ago, ohshitgorillas said:

To be honest, I'm less concerned about DSD performance since, while my DAC can technically do DSD, it's the PCM R-2R ladder where it really shines and I have more than enough processing power to run any PCM conversion. If my media server falls short of being able to do DSD512 on the most demanding filters, that's okay with me. That being said, I do wonder if it's worthwhile looking into reinstalling Ubuntu Server with the low latency kernel?

And remember that most demanding filters are not necessarily the technically best filters...

13 hours ago, ohshitgorillas said:

I am more concerned with getting NAA working on my RPi4/Pi2AES. Again, I can ping the device on the network, but it never shows up on the list of NAA endpoints.

It maybe not reachable on discovery. Or alternatively it doesn't have any available audio devices.

ohshitgorillas · January 11

11 hours ago, Miska said:

Not all settings in Embedded are in the web interface because Embedded is for hardware manufacturers building streamers. So the manufacturer knows their hardware and how optimally set up HQPlayer for their hardware. And end user purchasing such hardware who uses the web interface doesn't need to know about such details.

Well, what about for those of us who are running HQPlayer on a multi-purpose media server? I know my hardware but I'm not sure that HQPlayer is set up fully optimally yet. I'd like to be sure that mutli-core DSP is fully enabled. I'm also still considering using the low latency kernel.

11 hours ago, Miska said:

As help pages in the web interface. Help link is near top right corner of each corresponding web interface page.

Can't believe I never saw this, lol

11 hours ago, Miska said:

It maybe not reachable on discovery. Or alternatively it doesn't have any available audio devices.

I had to manually add the following to the config file on the NAA drive to get it working:

dtoverlay=hifiberry-digi-pro

NAA works now.

Miska · January 11

10 minutes ago, ohshitgorillas said:

Well, what about for those of us who are running HQPlayer on a multi-purpose media server?

Then those hardware related things are controlled by manually editing the hqplayerd.xml configuration file. This also goes for things like configuring inputs and also some output device options. It also gives more flexibility.

10 minutes ago, ohshitgorillas said:

I know my hardware but I'm not sure that HQPlayer is set up fully optimally yet. I'd like to be sure that mutli-core DSP is fully enabled. I'm also still considering using the low latency kernel.

Whether you get better performance with multicore DSP forced all-in, or in auto mode (default) depends on the particular hardware. So it usually takes some benchmarking to figure out. Auto mode likely performs better especially for cases where there are not so many physical cores.

10 minutes ago, ohshitgorillas said:
I had to manually add the following to the config file on the NAA drive to get it working:
dtoverlay=hifiberry-digi-pro
NAA works now.

OK, great!

dtoverlay means "device tree overlay", those are device tree snippets that can be added to the complete device tree. Which is Linux name for the thing that describes all existing hardware and their pin connections. When you add such line you tell the OS that such piece of hardware exists, what it is and how it is connected to the peripheral bus.

ohshitgorillas · January 11

5 minutes ago, Miska said:

Then those hardware related things are controlled by manually editing the hqplayerd.xml configuration file. This also goes for things like configuring inputs and also some output device options. It also gives more flexibility.

Whether you get better performance with multicore DSP forced all-in, or in auto mode (default) depends on the particular hardware. So it usually takes some benchmarking to figure out. Auto mode likely performs better especially for cases where there are not so many physical cores.

I have an i5-10600KF with twelve cores. Is that more than "not so many", or enough to justify enabling multi-core DSP forced all-in?

I'm also wondering if it's worth offloading to CUDA when just oversampling PCM to 4x?

Miska · January 11

12 minutes ago, ohshitgorillas said:

I have an i5-10600KF with twelve cores. Is that more than "not so many", or enough to justify enabling multi-core DSP forced all-in?

It has six physical cores. See here:

https://ark.intel.com/content/www/us/en/ark/products/199315/intel-core-i5-10600kf-processor-12m-cache-up-to-4-80-ghz.html

I have one generation newer i5-11600, but it is lower TDP 65W CPU. IIRC, it is cooled with Noctua NH-U9S cooler. And it can do DSD512...

Yours is K-series 95W TDP CPU and it should really be able to do DSD512 with EC-light modulators without any trouble. So if it cannot, I would look into why. Enough cooling? Fast enough RAM? BIOS settings OK?

You can try to force it to fully enabled, but it doesn't necessarily help. My old i5-7600T (35W TDP) is doing upsampling to DSD256 with generic build and multicore set to auto.

Baseline settings:

1x = poly-sinc-gauss-long

Nx = poly-sinc-gauss-hires-lp

Modulator = ASDM7EC-light

12 minutes ago, ohshitgorillas said:

I'm also wondering if it's worth offloading to CUDA when just oversampling PCM to 4x?

No, upsampling PCM to 4x (176.4/192k) can work on pretty much any CPU, even RPi4.

ohshitgorillas · January 11

6 minutes ago, Miska said:

It has six physical cores. See here:

https://ark.intel.com/content/www/us/en/ark/products/199315/intel-core-i5-10600kf-processor-12m-cache-up-to-4-80-ghz.html

I have one generation newer i5-11600, but it is lower TDP 65W CPU. IIRC, it is cooled with Noctua NH-U9S cooler. And it can do DSD512...

Yours is K-series 95W TDP CPU and it should really be able to do DSD512 with EC-light modulators without any trouble. So if it cannot, I would look into why. Enough cooling? Fast enough RAM? BIOS settings OK?

Baseline settings:

1x = poly-sinc-gauss-long

Nx = poly-sinc-gauss-hires-lp

Modulator = ASDM7EC-light

No, upsampling PCM to 4x (176.4/192k) can work on pretty much any CPU, even RPi4.

It can do DSD512 with EC-super with most filters except really demanding ones. I also use a Noctua NH-U9S cooler with two fans.

Regarding CUDA, is there any downside to leaving CUDA enabled and letting the GPU do part of the processing for PCM at 4x?

Also, could you please add Year to the HQPlayer Client, on the left-hand side of Play view, along with Artist/Album/Song?

Miska · January 11

1 minute ago, ohshitgorillas said:

Regarding CUDA, is there any downside to leaving CUDA enabled and letting the GPU do part of the processing for PCM at 4x?

No, there's no downside. It is just not necessary for this case.

1 minute ago, ohshitgorillas said:

Also, could you please add Year to the HQPlayer Client, on the left-hand side of Play view, along with Artist/Album/Song?

I can think about it. I have now unused "verbose info" flag that enables/disables the output information. I could overload this flag for the purpose so that for example year and genre are displayed when those output information is not shown. Composer and performer (lists) are already shown dynamically when they exist in the metadata, these could be also conditional on the verbose flag.

I would prefer to keep the text panel short enough that one wouldn't need to resort to scrolling it as much. More like single glance view. For next release there are already some new GUI elements on that column that take space of about one or two lines of text there.

ohshitgorillas · January 12

I am still getting drop-outs from Qobuz, streaming from HQPlayer Client to a NAA. These are rare and usually occur a few seconds before the end of a track, but have occurred in the middle of the track. This is the NAA behind a smart managed switch (TP Link TL-SG105E), which has no settings related to 802.3x flow control. The QoS setting is currently DSCP/802.1P Based. The device is a USBridge Sig running RoPieee, since the NAA OS won't run on the USBridge Sig on its own.

A few other suggestions or ideas, I apologize if this is out of line or not possible:

HQPlayer Client could auto-connect to the hqplayer instance on the network. Unless one has multiple instances of HQP on the network, which seems unlikely, or unless I'm missing something, which is far more likely, this feature basically always does the same thing and therefore is a prime candidate for automation. When resetting HQP, e.g. to change settings, Client could just wait and reconnect automatically.
It would be great to see some kind of library auto-update function, as most media servers these days do this. The fact that HQPDcontrol doesn't automatically sync its library with HQPlayer had me confused for like two weeks as to why some of my music wasn't showing up.

Miska · January 12

5 hours ago, ohshitgorillas said:

When resetting HQP, e.g. to change settings, Client could just wait and reconnect automatically.

This is quite unusual thing to happen. But it is essentially same as restarting HQPlayer.

5 hours ago, ohshitgorillas said:

It would be great to see some kind of library auto-update function, as most media servers these days do this.

On purpose, HQPlayer doesn't scan the storage behind your back. This is something you wouldn't want to happen unexpectedly. You know when you add/change content and you can then trigger rescan.

5 hours ago, ohshitgorillas said:

The fact that HQPDcontrol doesn't automatically sync its library with HQPlayer had me confused for like two weeks as to why some of my music wasn't showing up.

This is specific to HQPDcontrol which I cannot comment much about, since it is not my software.

But HQPlayer Client fetches the library data every time you connect to the server.

Miska · January 12

5 hours ago, ohshitgorillas said:

This is the NAA behind a smart managed switch (TP Link TL-SG105E), which has no settings related to 802.3x flow control.

According to it's manual yes it has, and according to the same manual it defaults to off. So first thing is to enable it for all ports. And you can also verify from the web interface that it actually becomes active on all involved ports (since it is negotiated at hardware level).

Miska · January 12

I've been testing BubbleUPnP beta for couple of days now with Tidal.

With these settings (everything didn't fit in a single screenshot properly, but one can figure out all relevant things anyway):

Several hours of playback with zero issues. Gapless, no proxy.

ohshitgorillas · January 14

On 1/12/2024 at 3:19 PM, Miska said:

According to it's manual yes it has, and according to the same manual it defaults to off. So first thing is to enable it for all ports. And you can also verify from the web interface that it actually becomes active on all involved ports (since it is negotiated at hardware level).

Got it, yeah, I overlooked that since I was looking for 802.3** rather than the words Flow Control.

I'm actually still getting drop outs near the end of songs even with Flow Control enabled on the switch. Usually, the drop-out occurs when HQPlayer Client appears to switch to the next track, which is slightly premature since the HQPlayer Client tracker is slightly ahead of the actual music. When it moves to the next track, there is a chance of the music dropping out.

Freewheel and prefetch are disabled.

There is another device in the path that may be affecting things. It's a GL iNet AX1800 wifi router (OpenWRT), although it's acting as a switch and wifi access point only rather than as a router. It sits about halfway between the router and the smart switch and connects both since I don't have a cable long enough to reach the switch directly from the router. I'll check its settings for anything like Flow Control or IGMP snooping.

Miska · January 14

8 minutes ago, ohshitgorillas said:

I'm actually still getting drop outs near the end of songs even with Flow Control enabled on the switch. Usually, the drop-out occurs when HQPlayer Client appears to switch to the next track, which is slightly premature since the HQPlayer Client tracker is slightly ahead of the actual music. When it moves to the next track, there is a chance of the music dropping out.

There is always a load spike when HQPlayer starts loading the next track. If the machine is on the edge of keeping up with the processing demands, this is the place where you usually flip it over and then you have a drop-out as result.

8 minutes ago, ohshitgorillas said:

There is another device in the path that may be affecting things. It's a GL iNet AX1800 wifi router (OpenWRT), although it's acting as a switch and wifi access point only rather than as a router. It sits about halfway between the router and the smart switch and connects both since I don't have a cable long enough to reach the switch directly from the router. I'll check its settings for anything like Flow Control or IGMP snooping.

802.3x is related to Ethernet standard, but not to WiFi which is a separate standard. So the type of flow control over WiFi is different. Of course WiFi can add some extra delay and unpredictability to the picture depending on conditions.

ohshitgorillas · January 14

11 minutes ago, Miska said:

802.3x is related to Ethernet standard, but not to WiFi which is a separate standard. So the type of flow control over WiFi is different. Of course WiFi can add some extra delay and unpredictability to the picture depending on conditions.

Sorry, this is a Wifi AP, but none of my music streamers use wifi at all, they are all wired directly into the network. So the fact that this does wifi is irrelevant.

I'm having a hard time navigating the OpenWRT settings, do you have any advice for how to set this up properly? That is, enabling IGMP snooping and flow control on OpenWRT?

11 minutes ago, Miska said:

There is always a load spike when HQPlayer starts loading the next track. If the machine is on the edge of keeping up with the processing demands, this is the place where you usually flip it over and then you have a drop-out as result.

I am using the sinc-M filter with NS9 dithering upsampling PCM to 4x. The PC should be far, far from the edge of keeping up with processing demands.

Miska · January 14

25 minutes ago, ohshitgorillas said:

I'm having a hard time navigating the OpenWRT settings, do you have any advice for how to set this up properly? That is, enabling IGMP snooping and flow control on OpenWRT?

I have not been using OpenWRT, so cannot help much on that. But I believe it has flow control enabled without extra hassle, as most NIC drivers do that by default. If your smart switch is connected to that one, you can see the flow control status from the switch settings page (same place where you enable it).

If your NAA discovery and such are working, you don't need to worry about IGMP Snooping.

25 minutes ago, ohshitgorillas said:

I am using the sinc-M filter with NS9 dithering upsampling PCM to 4x. The PC should be far, far from the edge of keeping up with processing demands.

Have you tried if using for example poly-sinc-ext3 (variant of the same filter as sinc-M) makes any difference? It can be actually heavier, but it has lower delay.

ohshitgorillas · January 14

27 minutes ago, Miska said:

Have you tried if using for example poly-sinc-ext3 (variant of the same filter as sinc-M) makes any difference? It can be actually heavier, but it has lower delay.

No, that still skips near the end of the track.

ohshitgorillas · January 14

Is there a way to add an album to the queue in HQPlayer while another album is still playing? Or do I have to use HQPDcontrol for that?

Miska · January 14

1 hour ago, ohshitgorillas said:

Is there a way to add an album to the queue in HQPlayer while another album is still playing? Or do I have to use HQPDcontrol for that?

HQPlayer Client at least allows you to do that. I'm not sure about HQPDcontrol.

2 hours ago, ohshitgorillas said:

No, that still skips near the end of the track.

You could try editing /etc/default/hqplayerd file and add there:

HQPLAYER_BUFFER_TIME=2500

Then do "systemctl restart hqplayerd" and see if it helps. You can also try values like 5000.

ohshitgorillas · January 24

I'm still having this issue where the music drops out around the end of the song... let me review so you don't have to go back thru the thread:

I am running HQPlayer daemonized on a dedicated headless media server (specs below)
- I am using PCM so I shouldn't have any problems running into the system's limits
When streaming from Qobuz via HQPlayer Client, the seek bar is slightly ahead of the actual music. When the seek bar arrives at the end of the song (prematurely) and moves to the next track, the music drops out for a second.
- This is unique to using Qobuz and HQPlayer Client, as it does not happen when streaming from my local library using HQPlayer Client, and it does not happen when using mConnect (but mConnect has other problems like sometimes forgetting to advance to the next song).
- This does not happen with every song, in fact most songs play fine
I have two NAA endpoints, one plugged directly into the router and the other behind a smart switch. Both devices have this problem.
I have tried setting HQPLAYER_BUFFER_TIME=2500 and =5000 in /etc/default/hqplayer but the problem has persisted.

Miska · January 24

2 hours ago, ohshitgorillas said:

When streaming from Qobuz via HQPlayer Client, the seek bar is slightly ahead of the actual music. When the seek bar arrives at the end of the song (prematurely) and moves to the next track, the music drops out for a second.

This is unique to using Qobuz and HQPlayer Client, as it does not happen when streaming from my local library using HQPlayer Client, and it does not happen when using mConnect (but mConnect has other problems like sometimes forgetting to advance to the next song).

Prefetch and freewheel enabled? (default)

What kind of internet connection? I've tested on my three internet connections (no issues). 500 Mbps fiber, 400 Mbps fiber and 1 Gbps 5G.

ohshitgorillas · January 24

1 hour ago, Miska said:

Prefetch and freewheel enabled? (default)

What kind of internet connection? I've tested on my three internet connections (no issues). 500 Mbps fiber, 400 Mbps fiber and 1 Gbps 5G.

Prefetch and freewheel are both disabled.

Internet connection is 1.2 Gbps cable.

Miska · January 24

4 minutes ago, ohshitgorillas said:

Prefetch and freewheel are both disabled.

It would help trying the possible combinations and see if it makes a difference.

But seems like the network access of next track causes the audio playback to drop-out. Either the network interface driver disables interrupts for too long time (typical for some WiFi and BT drivers). Or otherwise the audio output thread is not getting scheduled properly at that moment, despite the fact that it is set to high priority by HQPlayer.

ohshitgorillas · January 28

On 1/24/2024 at 2:40 PM, Miska said:

It would help trying the possible combinations and see if it makes a difference.

But seems like the network access of next track causes the audio playback to drop-out. Either the network interface driver disables interrupts for too long time (typical for some WiFi and BT drivers). Or otherwise the audio output thread is not getting scheduled properly at that moment, despite the fact that it is set to high priority by HQPlayer.

Enabling both prefetch and freewheel seems to have resolved the issue! I haven't gotten a drop-out in days, just perfect gapless playback.

AudioDoctor · January 29

I have been having this problem with JPLAY and Embedded as well. I just disabled Freewheel in the Embedded options and will see how it goes.

AudioDoctor · January 29

7 minutes ago, AudioDoctor said:

I have been having this problem with JPLAY and Embedded as well. I just disabled Freewheel in the Embedded options and will see how it goes.

not entirely. I still have a small skip at the beginning of each track.

BubbleUPnP > HQPlayer OS repeats the beginning of tracks

Recommended Posts

Link to comment

Link to comment

Link to comment

Link to comment

Link to comment

Link to comment

Link to comment

Link to comment

Link to comment

Link to comment

Link to comment

Link to comment

Link to comment

Link to comment

Link to comment

Link to comment

Link to comment

Link to comment

Link to comment

Link to comment

Link to comment

Link to comment

Link to comment

Link to comment

Link to comment

Create an account or sign in to comment

Create an account

Sign in