Set sample rate on usb capture card from gstreamer pipeline

[essamgouda97]

Hello, I am using the below script:

import time
import torch
from typing import Any
import numpy as np
import torchaudio



import gi
gi.require_version('Gst', '1.0')
from gstreamer import GstApp, Gst, GObject
GObject.threads_init()
Gst.init(None)

count = 0
frames = []

def on_buffer(sink: GstApp.AppSink) -> Gst.FlowReturn:
    """Callback on 'new-sample' signal"""

    global count
    global frames
    sample = sink.emit("pull-sample")  # Gst.Sample
    

    if isinstance(sample, Gst.Sample):
        buffer = sample.get_buffer()
        buffer_size = buffer.get_size()
        data = buffer.extract_dup(0, buffer_size)

        #format, layout, rate, channels, channel-mask,
        if count == 1000:
            print(count)
            frames = torch.stack(frames).view(-1).unsqueeze(0)
            torchaudio.save("test.wav",frames,44100)
            frames = []
            count = 0
        else:
            frames.append(torch.from_numpy(np.fromstring(data, dtype=np.float32)))
            count+=1

        return Gst.FlowReturn.OK

    return Gst.FlowReturn.ERROR

#alsasrc device=hw:MS2109,0, pulsesrc device=alsa_input.usb-MACROSILICON_USB_Video-02.analog-stereo
command = "alsasrc device=hw:MS2109,0 ! appsink emit-signals=True drop=True"

pipeline = Gst.parse_launch(command)
appsink = pipeline.children[0]  # get AppSink
appsink.connect("new-sample", on_buffer)

pipeline.set_state(Gst.State.PLAYING)
time.sleep(200)

To capture audio data using appsink and save it using torchaudio. I have tried both pulsesrc and alsasrc but when I save the audio it sounds very choppy. My theory is that the sample rate isn’t aligned so my question is: how can I set the sample rate in this pipeline alsasrc device=hw:MS2109,0 ! appsink emit-signals=True drop=True. I am also open for any suggestions to improve the process in general, I just need to capture audio data and convert it to pytorch tensors so I can pass it to a pytorch model.

Edit 1: I would like to have a sample rate of 44.1k.
I have edited /etc/asound.conf and added:

pcm.!default {
        type plug
        slave {
                pcm "hw:MS2109,0"
                rate 44100
        }
        hint.description "default Soundcard"
}

in order to change the sample rate but I have no way of making sure that it is uing the correct sample rate.

Edit 2: I have changed the pipeline to:

"pulsesrc device=alsa_input.usb-MACROSILICON_USB_Video-02.analog-stereo ! audioconvert ! tee name=audioTee \
    audioTee. ! queue ! wavenc ! filesink location=test2.wav \
    audioTee. ! queue ! appsink emit-signals=True drop=True"

So I can make sure that whether the problem from the soundcard itself or is it the way I am handling the captured data after and using filesink I save the data into the file as expected and the sound is crystal clear but when I save the data using torchaudio from the line: torchaudio.save("test.wav",frames,44100) the generated audio file is very noisy.

[DaneLLL]

Hi,
Please make sure you can record good wav file in arecord command. And then you can apply sample rate to caps like:

alsasrc ! audio/x-raw,rate=48000 ! ...

Please check alsasrc for more information.