High CPU usage after speaker.Init() called

[duysqubix]

First of all, thanks for Beep and the excellent documentation.

I've written a doorbell in Go that listens to an MQTT channel and then plays a ding-dong noise when a button is pressed. I can deploy the code to multiple hosts on different platforms, which is the appeal of using Go for this.

My code runs as a systemd unit, starts up and initialises the sound streams using Beep's speaker package. However, I've noticed that the baseline CPU usages is pretty high - I have two wav files open and this is using ~15% CPU - even on a puny Celeron box this seems excessive. I've written some minimal code to reproduce this:

package main

import (
    "os"
    "github.com/faiface/beep"
    "github.com/faiface/beep/wav"
    "github.com/faiface/beep/speaker"    
    "time"
)

func main() {
    path := os.Args[1]

    f, _ := os.Open(path)

    streamer, format, _ := wav.Decode(f)    

    speaker.Init(format.SampleRate, format.SampleRate.N(time.Second/100))
    time.Sleep(30 * time.Second)
    done := make(chan bool)
    speaker.Play(beep.Seq(streamer, beep.Callback(func() {
        done <- true
    })))
    <-done

}

Save the snippet above as play.go and then run with go run play.go <wavfile>. During the time.Sleep (after init, but before the sound is actually played) you can see the usage in top.

I can reduce the CPU usage by reducing the sample rate in the Init call - with time.Second it was down to about 1% - but that still seems high since it should be completely idle. I also tried using buffering as per the wiki, but that didn't make any difference. Did I miss something here? I'm using go1.18.1.

Original issue: faiface/beep#159

[MarkKremer]

As mentioned in the original issue, I expect the speaker.Init was called with an invalid sample rate. I added an item to the collection issue to improve error checking.

Feel free to re-open the issue if this wasn't the problem.

[MarkKremer]

This got a new reply from the original author in the faiface/beep repo:

Sorry to comment on a closed issue and to take such a long time to get back to this: I tried this:

package main

import (
    "fmt"
    "os"
    "github.com/faiface/beep"
    "github.com/faiface/beep/wav"
    "github.com/faiface/beep/speaker"    
    "time"
)

func main() {
    path := os.Args[1]

    f, err := os.Open(path)
    if (err != nil) {
        fmt.Println(err)
    }

    streamer, format, err := wav.Decode(f)    
    if (err != nil) {
        fmt.Println(err)
    }

    fmt.Println(format.SampleRate)

    speaker.Init(format.SampleRate, format.SampleRate.N(time.Second/100))
    time.Sleep(300 * time.Second)
    done := make(chan bool)
    speaker.Play(beep.Seq(streamer, beep.Callback(func() {
        done <- true
    })))
    <-done

 }

Neither the os.Open() nor the wav.Decode() return an error. The SampleRate is reported as 44100, which seems about right. Is there anything else I can try?

[MarkKremer]

Some things to try:

• upgrade to the latest version of gopxl/beep. Some changes have been made to the speaker and it uses a newer version of Oto.
• pick a bigger buffersize. 1/100th of a second is too small. 1/60th is stretching it and 1/30th is more reasonable on most systems.

Which OS are you running?

Related issue: #137

[psaffrey]

Thanks @MarkKremer and I'm glad to see that this project lives on in another repo :)

I tried updating to gopxl/beep/[email protected] and I also increased the buffer size to time.Second/10, as in the "Hello, Beep!" example. The CPU usage is still around 15% and the pulseaudio CPU usage is also high. I got this with top -b -n 2 -d 0.2 -p 1031,4650:

    PID USER      PR  NI    VIRT    RES    SHR S  %CPU  %MEM     TIME+ COMMAND
   4650 psaffrey  20   0 1495404  10136   6708 S  15.0   0.1   0:01.41 play_minimal
   1031 psaffrey   9 -11 2472696  28432  21640 S  10.0   0.4   0:46.68 pulseaudio

I'm running Ubuntu 22.04.

[MarkKremer]

Could you profile this for me? See #137 (comment) but put the profiling code above your own example code that is giving you the high load and share both the profiler output and the code here.

I don't have time right now but I'll try to look into it within a couple weeks. Although from a quick look at the profile in the other issue it wasn't clear to me yet how it can be improved.