Reading audio data from an ALSA buffer to a numpy array

Question

I'm trying to pull data from an ALSA buffer in order to generate count noise from a mic. However, when I try to convert the data to an array I get an incorrect result.

Below is part of my code:

#!/usr/bin/env python

from __future__ import print_function

import alsaaudio
import numpy

card = 'default'
buf =  [64]
numpy.set_printoptions(threshold=numpy.inf)

stream = alsaaudio.PCM(alsaaudio.PCM_CAPTURE, alsaaudio.PCM_NORMAL, card)
stream.setchannels(1)
stream.setrate(44100)
stream.setformat(alsaaudio.PCM_FORMAT_S16_LE)
stream.setperiodsize(64)
def listen():
    print("Listening")
    while True:
        try:
            l, data = stream.read()

            f = open('test.raw', 'wb')
            if l:
                 f.write(data)
                 f.close()
        except IOError, e:
            error_count += 1
            print(" (%d) Error recording: %s" % (error_count, e))
        else:
            decoded_block = numpy.frombuffer(data, dtype='i2' )
            print('Array PCM: \n',decoded_block)

            return 0

listen()

Answer 1

hexdump -d displays the contents as unsigned 16bit integers, whereas you are converting it to a signed int16 numpy array.

Try converting to a dtype of 'u2' (or equivalently, np.uint16 ) and you'll see that the output matches that of hexdump :

print(np.array([-7, -9, -10, -6, -2], dtype=np.uint16))
# [65529 65527 65526 65530 65534]