Why does not program cost 1 sec when it plays 44100 samples with 44100 sample rate by using PyAudio?

Question

import pyaudio
import numpy as np
import time

RATE=44100
pa = pyaudio.PyAudio()
stream = pa.open(format=pyaudio.paFloat32,
                                channels=1,
                                rate=RATE,
                                output=True)
t = time.time()
output = np.ones(44100)*100
stream.write(output)
print(time.time()-t)

This is a test code. With 44100 sample rate, when playing 44100 samples, the time cost should be 1s. However, the output is not. Why does it happen? I am looking forward to someone's answer. Thank you very much.

And what is the output? – Roland Smith Aug 16 '17 at 13:15 — Roland Smith, Aug 16 '17 at 13:15

Peter Schneider · Accepted Answer · 2017-08-16T13:56:24.950

2

Disclaimer: untested, just looked into the API.

stream.write() expects a string of a defined format (here: 32bit-floatingpoint). But you pass a handle to a numpy array. PyAudio (probably) defaults to convert the passed value into a string. The string representation of your output without further modification is:

array([ 100.,  100.,  100., ...,  100.,  100.,  100.])

or 42 bytes or 2.3ms playtime.

Try stream.write('\0\0\0\0'*44100). To use np.ones() or numpy-arrays in general you have to use the correct datatype (you defined the format to be paFloat32):

output = np.ones(44100, dtype=np.float32).tobytes()

edited Aug 16 '17 at 13:56

answered Aug 16 '17 at 13:04

Peter Schneider

1,683
12
31

I added a code output.tostring() before stream.write(). But the result shows that it costed about 2s. – Roger Aug 16 '17 at 13:37
Thats because numpy defaults to double (64bit) if the datatype isn't defined. Updated answer... – Peter Schneider Aug 16 '17 at 13:57
May I ask one more question? The details are in my latest question. Thank you very much. – Roger Aug 16 '17 at 14:36

Why does not program cost 1 sec when it plays 44100 samples with 44100 sample rate by using PyAudio?

1 Answers1