[audio-server] Audio issues using alsa playback backend (drain error + sound is cut)

[carsten19]

What's the issue

I just upgraded my system (Raspi 4 with Stretch packages) to 0.64. Unfortunately the error
ALSA function 'snd_pcm_drain' failed with error 'EIO: I/O error'
is back. None of the known workarounds fix it.
Once I downgraded to 0.63.3 the problem disappears.

Expected
Same behavior as 1.2.2

Version

1.3.0 (0.64.0)

Environment

• Raspberry Pi 4 - Debian Stretch
• Sound card - Raspberry pi sound card

---

[workaround edit by @cpoisson ]

Workaround 1 - Back to 0.63.3

Downgrade snips-tts and snips-audio-server to 1.2.2 (0.63.3)

sudo apt install snips-audio-server=0.63.3
sudo apt install snips-tts=0.63.3

Workaround 2 - Back to portaudio

Switch back to portaudio for playback.

• Add portaudio_playback = "default" to the [snips-audio-server] section of your /etc/snips.toml configuration file.
• sudo systemctl restart snips-audio-server

Workaround 3 - bcm2835

If you are using the pi 3 or pi 4 bcm2835 embedded sound card.

• Switch back to portaudio
• Use the following configuration for playback

pcm.!default {
    type asym
    playback.pcm {
        type plug
        slave.pcm "hw:0,0"
        slave.rate 48000  # Does not work if this is removed.
    }
    capture.pcm {
        type plug
        slave.pcm "hw:1,0"   # Configure the correct hw:Id,SubId for your capture device
    }
}

Workaround 4 : Delegate to dmix and dsnoop

Delegate to dmix to make sure that the sound is played well using alsa

pcm.!default {
    type asym
    playback.pcm "playback"
    capture.pcm "capture"
}

pcm.playback {
    type plug
    slave.pcm "dmixed"
}

pcm.capture {
    type plug
    slave.pcm "array"
}

pcm.dmixed {
    type dmix
    ipc_key 555555
    ipc_key_add_uid yes
    ipc_perm 0666
    slave.pcm "hw:1,0"  # Set the correct hw:Id,SubId for your playback device
}

pcm.array {
    type dsnoop
    ipc_key 666666
    ipc_key_add_uid yes
    ipc_perm 0666
    slave.pcm "hw:1,0"  # Set the correct hw:Id,SubId for your capture device
}

This configuration has been tested on the following devices

Audio device	Playback	Capture	Details
bcm2835 (raspberry pi)	❌		Does not work. playback is stuck waiting for /dev/snd/timer (known issue)
Jabra PHS002W	✅	✅
UGREEN USB Sound card	✅	✅
PS Eye Camera		✅
Respeaker PiHAT 2 mics	✅	✅
Respeaker 6 Microphones	❌	✅	Playback does not work. The default configuration installed by the drivers works correctly tough

Related Issue

• [audio-server] Output sound like TTS is cut quickly when it starts playing for some configurations #201 : Sound partial playback (duplicated)
• [Platform][Audio-Server][PortAudio] Cannot play sound sampled at 22050Hz when playback is set at 16000Hz #148 : Other portaudio issue (22050 samples/s issue)

[cpoisson]

Thanks for your ticket @cburghardt,

Quick question, Which sound card are you using as your main sound playback interface?

Can you post as well your asound.conf configuration.

Many thanks for your feedbacks

[carsten19]

I use the onboard sound card of the raspberry.

pcm.jack {                  
        type hw             
        card 0              
        device 0            
        rate 16000          
} 

pcm.!default {
  type asym
   playback.pcm {
     type plug
     slave.pcm "jack"
   }
   capture.pcm {
     type plug
     slave.pcm "hw:1,0"
   }
}

[cpoisson]

Ok,

So the issue is indeed related to #201

Using alsa, the ring buffer parameters setup done by the audio server is incorrect, and seems to induce multiple issues:

• snips-audio-server is deadlocked (probably waiting for the snd_pcm_drain (all audio samples are played) and need a restart.
• sound is sometimes cut (work / does not work, depends on your luck + sample rate)

This issue seems to appears with different behavior depending on your sound card.

Pi 3 Embedded sound card

pi@snips-pi-qa-assistant:~ $ cat /proc/asound/card0/pcm0p/sub0/hw_params
access: MMAP_INTERLEAVED
format: S16_LE
subformat: STD
channels: 1
rate: 48000 (48000/1)
period_size: 512
buffer_size: 65536

Jabra

pi@snips-pi-qa-assistant:~ $ cat /proc/asound/card1/pcm0p/sub0/hw_params
access: MMAP_INTERLEAVED
format: S16_LE
subformat: STD
channels: 2
rate: 48000 (48000/1)
period_size: 256
buffer_size: 262144

Typical parameters induced by aplay

• period_size: 6000; buffer_size: 24000 @ 48000Hz
• period_size: 4000; buffer_size: 16000 @ 22050Hz
• period_size: 2000; buffer_size: 8000 @ 16000Hz

What seems to work
Using the following configuration and portaudio backend. Things seems to work again.

pcm.!default {
    type asym
    playback.pcm {
        type plug
        slave.pcm "hw:0,0" # the pi3 sound card
        slave.rate 48000 # to avoid port audio sampling issues

    }
    capture.pcm {
        type plug
        slave.pcm "hw:2,0" # A usb microphone plugged on the my raspberry pi.
    }
}

Which give me the following hw parameters when playing a sound sample with the audio-server.

pi@snips-pi-qa-assistant:~ $ cat /proc/asound/card0/pcm0p/sub0/hw_params
access: MMAP_INTERLEAVED
format: S16_LE
subformat: STD
channels: 2
rate: 48000 (48000/1)
period_size: 836
buffer_size: 4184

The buffer size is back to a reasonable value. The sound issues disappears.

(I found some documentation for the one interested)

[cpoisson]

@marcteyssier, @cburghardt

I gave a try to your configuration (coming from the forum I suppose), using portaudio.

snips-audio-server -v --portaudio_playback=default

Seems to work correclty I don't see any errors.

Can you patch your /etc/snips.toml with portaudio_playback = "default" under [snips-audio-server] section, to give a shot to this workaround (1.3.0 only) ?

[carsten19]

So I upgraded to 0.64 and applied the changes. Result:
caused by: an error occured in the audio pipeline: ErrorMessage { msg: "Couldn't find a PortAudio device with the name default" }

[cpoisson]

Ok, that's another one to figure out.

I also tried to delegate things to dmix. Works great with the piHAT 2 (provided the user using the sound is in the audio group)

But as soon as I switch on the raspberry pi even aplay deadlock. After an inspection with strace I think I stumbled on this known issue

[marcteyssier]

Hi, this work for me with my respeaker mic 4 array.
I heard all the sentence.

Thank you for your help

[cpoisson]

Regarding the bcm2835 (raspi sound card) it is really impossible to use it with dmix on my side. The thing just stay stuck.

Otherwise the following asound.conf configuration works perfectly on the hw available on my desk using the default alsa playback.

/etc/asound.conf

pcm.!default {
    type asym
    playback.pcm "playback"
    capture.pcm "capture"
}

pcm.playback {
    type plug
    slave.pcm "dmixed"
}

pcm.capture {
    type plug
    slave.pcm "array"
}

pcm.dmixed {
    type dmix
    ipc_key 555555
    ipc_key_add_uid yes
    ipc_perm 0666
    slave.pcm "hw:5,0"
}

pcm.array {
    type dsnoop
    ipc_key 666666
    ipc_key_add_uid yes
    ipc_perm 0666
    slave.pcm "hw:5,0"
}

Hardware tested

Audio device	Playback	Capture	Details
bcm2835 (raspberry pi)	❌		playback is stuck waiting for /dev/snd/timer (known issue)
Jabra PHS002W	✅	✅
UGREEN USB Sound card	✅	✅
PS Eye Camera		✅
Respeaker PiHAT 2 mics	✅	✅