PyAudio behavior during time.sleep()

Question

I'm trying to understand something I'm seeing in PyAudio. I'm trying to take a 200ms sample of audio, wait a few seconds, and then take three more 200ms samples of audio. Consider this code:

import pyaudio
import time

p = pyaudio.PyAudio()
chunk = 1024
FORMAT = pyaudio.paInt16
CHANNELS = 1 #chan
RATE = 11025
stream = p.open(format = FORMAT,
            channels = CHANNELS,
            rate = RATE,
            input = True,
            output = False,
            frames_per_buffer = chunk,
            input_device_index = 0,
            output_device_index = 0)

def record(seconds):
all = []
for i in range(0, int(RATE / chunk * seconds)):
    data = stream.read(chunk)

    all.append(data)
data = ''.join(all)
return data

#record 200ms of sound
print "pre-record 1 " + str(time.time())
data = record(.2)
print "post-record 1 " + str(time.time())

#sleep for one second
#time.sleep(1)

#record 200ms of sound
print "pre-record 2 " + str(time.time())
data = record(.2)
print "post-record 2 " + str(time.time())

print "pre-record 3 " + str(time.time())
data = record(.2)
print "post-record 3 " + str(time.time())

print "pre-record 4 " + str(time.time())
data = record(.2)
print "post-record 4 " + str(time.time())

If I run this "as-is" (i.e. with time.sleep() commented out), I get this, which makes sense:

pre-record 1 1357526364.46
post-record 1 1357526364.67
pre-record 2 1357526364.67
post-record 2 1357526364.86
pre-record 3 1357526364.86
post-record 3 1357526365.03
pre-record 4 1357526365.03
post-record 4 1357526365.22

If I then uncomment the time.sleep() line to add a one second delay between the first and second recording, I get this:

pre-record 1 1357525897.09
post-record 1 1357525897.28
pre-record 2 1357525898.28
post-record 2 1357525898.28
pre-record 3 1357525898.28
post-record 3 1357525898.28
pre-record 4 1357525898.28
post-record 4 1357525898.47

Notice that while there is a one-second delay in the timestamps between the first and second recordings, there is no delay at all between recordings 2 and 3, and they appear to have been taken in zero time (although they actually do contain 200ms of data). Recording four appears to have taken 200ms to get, but was started at the same time as recordings 2 and 3.

If I replace time.sleep(1) with record(1) (and not save/use the data from the 1-second recording), the program behaves as I expect:

pre-record 1 1357526802.57
post-record 1 1357526802.77
pre-record 2 1357526803.69
post-record 2 1357526803.88
pre-record 3 1357526803.88
post-record 3 1357526804.06
pre-record 4 1357526804.06
post-record 4 1357526804.25

So it would appear that even during the time.sleep(1), a buffer somewhere is still being fed audio, and when I call the record function after the sleep, it's grabbing the audio from that buffer, which is not the audio I want. I need the audio after the sleep, not during it. Can someone help me understand PyAudio's interaction with whatever buffer is there, and is there a way to better know at what time my audio was actually captured?

score 0 · Accepted Answer · answered Jan 07 '13 at 03:42

0

Audio device will add RATE samples per second to the buffer continually, and your code read from this buffer.

So you can't sleep(1) without reading from buffer. You can write some code to skip samples:

def skip(seconds):
    samples = int(seconds * RATE)
    count = 0
    while count < samples:
        stream.read(chunk)
        count += chunk
        time.sleep(0.01)

answered Jan 07 '13 at 03:42

HYRY

94,853
25
187
187

Thanks, this is very helpful. Do you know if the buffer has a certain size? If you look at the second example above, it looks like recording 4 actually took 200ms to complete, which would seem to indicate that maybe the buffer can only hold about 400 ms worth of data at this RATE? – aaknitt Jan 07 '13 at 14:36
1

To partially answer my own question, there is a good discussion on the buffer size here: http://www.portaudio.com/docs/latency.html It looks like there is a PortAudio function `Pa_GetMinNumBuffers( int framesPerBuffer, double sampleRate )` that will return the buffer size but I haven't found that in PyAudio yet. However, by calling the `get_read_available()` function I can get some idea on how the buffer is behaving. – aaknitt Jan 07 '13 at 19:41
Also, it appears that calling `stream.stop_stream()` prior to `time.sleep(1)` will stop the audio device from adding to the buffer, and then calling `stream.start_stream()` after the sleep will restart adding to the buffer. The timing seems to work out by doing this. – aaknitt Jan 07 '13 at 19:48

PyAudio behavior during time.sleep()

1 Answers1