spatialaudio / jackclient-python

🂻 JACK Audio Connection Kit (JACK) Client for Python :snake:
https://jackclient-python.readthedocs.io/
MIT License
131 stars 26 forks source link

Real-time plotting example #106

Open vyhyb opened 3 years ago

vyhyb commented 3 years ago

Hi, thank you again for your great library. I have a bit problems with real time plotting of gathered audio data (and its spectrum). I guess it is probably some threading issue, is it? Don't you know how to deal with this? I've tried several ways but I don't have any luck with updating the graph. Thanks in advance!

Code here:

import numpy as np
import queue
import sys
import threading

buffersize = 3
clientname = "ImpedanceTubeTest2"
rec_time = 10
q = queue.Queue(maxsize=buffersize*10)
global rec_end
rec_end = False
event = threading.Event()

def print_error(*args):
    print(*args, file=sys.stderr)

def xrun(delay):
    print_error("An xrun occured, increase JACK's period size?")

def shutdown(status, reason):
    print_error('JACK shutdown!')
    print_error('status:', status)
    print_error('reason:', reason)
    event.set()

def stop_callback(msg=''):
    if msg:
        print_error(msg)
    event.set()
    raise jack.CallbackExit

def process(frames):
    if frames != blocksize:
        stop_callback('blocksize must not be changed, I quit!')
    if rec_end:
        stop_callback()  # Recording is finished
    try:
        q.put_nowait(client.inports[0].get_array()[:])
    except (queue.Full):
        print("Full Queue")
        stop_callback()

from IPython.display import clear_output
from time import sleep
from scipy.signal import stft
import matplotlib.pyplot as plt
from matplotlib.animation import FuncAnimation
size = 8
olap = 2
olap_inv = size - olap

global fig, ax
fig, ax = plt.subplots()

def animate(i, spectrum):
    line.set_ydata(spectrum)  # update the data.
    fig.canvas.draw()
    return line,

try:
    import jack

    client = jack.Client(clientname, no_start_server=True)
    blocksize = client.blocksize
    samplerate = client.samplerate
    FFT_size = size * blocksize
    overlap = olap * blocksize
    overlap_inv = olap_inv*blocksize
    client.set_xrun_callback(xrun)
    client.set_shutdown_callback(shutdown)
    client.set_process_callback(process)
    client.inports.register('in_1')
    data = np.zeros(FFT_size)
    init = stft(data, fs= samplerate, nperseg=FFT_size
    spectrum = init[2].T[0].real
    line, = ax.plot(init[0],spectrum)

    ani = FuncAnimation(
        fig, animate, interval=1, fargs=[spectrum])
    plt.show(block=False)
    sleep(2)
    with client:
        target_ports = client.get_ports(
            is_physical=True, is_output=True, is_audio=True)

        client.inports[0].connect(target_ports[0])
        timeout = blocksize * buffersize / samplerate
        for i in range(olap):
            data[i*blocksize:(i+1)*blocksize] = q.get(timeout=timeout)
        while True:
            try:
                for i in range(olap,size):
                    data[i*blocksize:(i+1)*blocksize] = q.get(timeout=timeout)

                stft_data = stft(data, fs= samplerate, nperseg=FFT_size)
                spectrum = stft_data[2]*np.conj(stft_data[2])/FFT_size
                # print(spectrum.max()) #peak output for reference
                # clear_output(wait=True) #reference
                data[0:overlap] = data[overlap_inv:FFT_size]
            except KeyboardInterrupt:
                print("Interrupted by user")
        rec_end = True
        event.wait()  # # # Wait until recording is finished
except (queue.Empty):
    # A timeout occured, i.e. there was an error in the callback
    print("Empty Queue")
mgeier commented 3 years ago

Did you have a look at https://github.com/spatialaudio/python-sounddevice/blob/master/examples/plot_input.py?

I know it's not using the jack module and it's not calculating the spectrum, but it should be relatively easy to change that.

Regarding your code, I'm not sure if spectrum can be passed to the animate() function like this. And it looks like spectrum may be accessed by multiple threads at the same time?

Is it possible to move your calculations (including reading from the q) into the animate() function?

mgeier commented 3 years ago

@vyhyb Any news on this?

vyhyb commented 3 years ago

Yes, here is the final code. Thank you for your help.

import numpy as np
import queue
import sys
import threading

buffersize = 3
clientname = "ImpedanceTubeTest2"
q = queue.Queue(maxsize=buffersize*100)
global rec_end
rec_end = False
event = threading.Event()

def print_error(*args):
    print(*args, file=sys.stderr)

def xrun(delay):
    print_error("An xrun occured, increase JACK's period size?")

def shutdown(status, reason):
    print_error('JACK shutdown!')
    print_error('status:', status)
    print_error('reason:', reason)
    event.set()

def stop_callback(msg=''):
    if msg:
        print_error(msg)
    event.set()
    raise jack.CallbackExit

def process(frames):
    if frames != blocksize:
        stop_callback('blocksize must not be changed, I quit!')
    if rec_end:
        stop_callback()  # Recording is finished
    try:
        q.put_nowait(client.inports[0].get_array()[:])
    except (queue.Full):
        print("Full Queue")
        stop_callback()

from IPython.display import clear_output
from time import sleep
from scipy.signal import stft
import matplotlib.pyplot as plt
from matplotlib.animation import FuncAnimation
import matplotlib as mpl

mpl.use("Qt5Agg")

size = 4
olap = 0
olap_inv = size - olap

global fig, ax
fig, ax = plt.subplots()

sound = None
log_spec = None

def animate(i):
    global sound, spectrum, stft_data, log_spec
    while True:
        try:
            data = q.get_nowait()
        except queue.Empty:
            break
        shift = len(data)
        sound = np.roll(sound, -shift, axis=0)
        sound[-shift:] = data
    # arr = plt.mlab.specgram(sound, NFFT=blocksize, Fs=48000)[0]
    # print(arr.shape)
    # im.set_array(arr)
    stft_data = stft(sound, fs= samplerate, nperseg=FFT_size)
    spectrum = stft_data[2]*np.conj(stft_data[2])
    # print(spectrum.max())
    log_spec = (5*log_spec + 10*np.log10(spectrum.mean(axis=1)))/6
    line.set_ydata(log_spec)
    return line,

try:
    import jack

    client = jack.Client(clientname, no_start_server=True)

    blocksize = client.blocksize
    samplerate = client.samplerate
    FFT_size = size * blocksize
    overlap = olap * blocksize
    overlap_inv = olap_inv*blocksize

    client.set_xrun_callback(xrun)
    client.set_shutdown_callback(shutdown)
    client.set_process_callback(process)
    client.inports.register('in_1')

    timeout = blocksize * buffersize / samplerate
    sound = np.ones(blocksize*32)
    stft_data = stft(sound, fs= samplerate, nperseg=FFT_size)
    spectrum = stft_data[2]*np.conj(stft_data[2])
    log_spec = 10*np.log10(spectrum.mean(axis=1))
    line, = ax.plot(stft_data[0], log_spec)
    # line.set_ydata(spectrum.T[0])
    # fig = plt.figure(figsize=(10,5))
    # arr, freqs = plt.mlab.specgram(sound, NFFT=blocksize, Fs=samplerate)[0:2]
    # im = plt.imshow(arr, animated=True)
    ax.set_xlim(20, samplerate/2)
    ax.set_ylim(-140, 0)
    ax.set_xscale('log')
    ani = FuncAnimation(
        fig, animate, interval=timeout, blit=True)

    with client:
        target_ports = client.get_ports(
            is_physical=True, is_output=True, is_audio=True)
        client.inports[0].connect(target_ports[0])
        # plt.ion()
        plt.show()
        rec_end = True
        event.wait()  # # # Wait until recording is finished
except (queue.Empty):
    # A timeout occured, i.e. there was an error in the callback
    print("Empty Queue")
mgeier commented 3 years ago

Cool, thanks!

Would you like to contribute this to the examples?