[Raspi4; Introduction to Sound] Stable recording of sound input with python ♪

I've covered sound input several times, but this time I've been able to input and output stably with Rasipi4, so I decided to summarize it. The important thing is as shown in Reference 1. Last year, I was playing with something like Reference 2. I'm thinking of doing something a little different this year. 【reference】 1.PyAudio Input overflowed 2. [Scipy] I played with FFT, STFT and wavelet transform ♬ ～ ⑦ Real-time spectrogram; Speed up

The point

In other words, if you write the following magic when reading, overflow will not occur and you can record endlessly.

• This article is a story that this was solved even with Rasipi4.

# -*- coding:utf-8 -*-
import pyaudio
import matplotlib.pyplot as plt
import numpy as np
import wave
import struct

Save the read sound as a wav file with the following function.

def savewav(sig,sk):
    RATE = 44100 #Sampling frequency
    #Sine wave-Convert from 32768 to 32767 integer value(to signed 16bit pcm)
    swav = [(int(32767*x)) for x in sig] #32767
    #Binary
    binwave = struct.pack("h" * len(swav), *swav)
    #Export sine wave as wav file
    w = wave.Wave_write("./wine/"+str(sk)+".wav")
    params = (1, 2, RATE, len(binwave), 'NONE', 'not compressed')
    w.setparams(params)
    w.writeframes(binwave)
    w.close()

Sound input / output is set to True. Since voice (sentence) is input, record a long time (102400/44100 = 2.3sec).

RATE=44100
p=pyaudio.PyAudio()
N=100
CHUNK=1024*N
stream=p.open(format = pyaudio.paInt16,
        channels = 1,
        rate = RATE,
        frames_per_buffer = CHUNK,
        input = True,
        output = True) #Set input and output to True at the same time

And finally, continuous recording has started. Exception_on_overflow is set to False. sig is divided by 32768.

sk=0
while stream.is_active():
    input = stream.read(CHUNK, exception_on_overflow = False)
    print(len(input))
    sig =[]
    sig = np.frombuffer(input, dtype="int16") / 32768
    savewav(sig,sk)
    fig, (ax1,ax2) = plt.subplots(2,1,figsize=(1.6180 * 4, 4*2))
    lns1=ax1.plot(sig[0:1024] ,".-",color="red")
    ax1.set_xticks(np.linspace(0, 882, 3))
    ax1.set_ylabel("sig0")
    ax1.set_title('short plot')
    lns2=ax2.plot(sig[0:CHUNK], "-",color="blue")
    ax2.set_xticks(np.linspace(0, 44100*2, 5))
    ax2.set_ylabel("sig1")
    ax2.set_title('long plot')
    ax1.grid()
    ax2.grid()
    plt.pause(0.5)
    plt.savefig("./wine/sound_{}.png ".format(sk))
    plt.close()
    sk+=1
    output = stream.write(input)

Ah, uh, eh. .. Continuous measurement result of x-axis memory; 44100 = 1sec

Summary

・ Continuous measurement is now possible

・ I want to link R-python with sound.