record audio with processing python

1from sys import byteorder
2from array import array
3from struct import pack
4
5import pyaudio
6import wave
7
8THRESHOLD = 500
9CHUNK_SIZE = 1024
10FORMAT = pyaudio.paInt16
11RATE = 44100
12
13def is_silent(snd_data):
14    "Returns 'True' if below the 'silent' threshold"
15    return max(snd_data) < THRESHOLD
16
17def normalize(snd_data):
18    "Average the volume out"
19    MAXIMUM = 16384
20    times = float(MAXIMUM)/max(abs(i) for i in snd_data)
21
22    r = array('h')
23    for i in snd_data:
24        r.append(int(i*times))
25    return r
26
27def trim(snd_data):
28    "Trim the blank spots at the start and end"
29    def _trim(snd_data):
30        snd_started = False
31        r = array('h')
32
33        for i in snd_data:
34            if not snd_started and abs(i)>THRESHOLD:
35                snd_started = True
36                r.append(i)
37
38            elif snd_started:
39                r.append(i)
40        return r
41
42    # Trim to the left
43    snd_data = _trim(snd_data)
44
45    # Trim to the right
46    snd_data.reverse()
47    snd_data = _trim(snd_data)
48    snd_data.reverse()
49    return snd_data
50
51def add_silence(snd_data, seconds):
52    "Add silence to the start and end of 'snd_data' of length 'seconds' (float)"
53    silence = [0] * int(seconds * RATE)
54    r = array('h', silence)
55    r.extend(snd_data)
56    r.extend(silence)
57    return r
58
59def record():
60    """
61    Record a word or words from the microphone and 
62    return the data as an array of signed shorts.
63
64    Normalizes the audio, trims silence from the 
65    start and end, and pads with 0.5 seconds of 
66    blank sound to make sure VLC et al can play 
67    it without getting chopped off.
68    """
69    p = pyaudio.PyAudio()
70    stream = p.open(format=FORMAT, channels=1, rate=RATE,
71        input=True, output=True,
72        frames_per_buffer=CHUNK_SIZE)
73
74    num_silent = 0
75    snd_started = False
76
77    r = array('h')
78
79    while 1:
80        # little endian, signed short
81        snd_data = array('h', stream.read(CHUNK_SIZE))
82        if byteorder == 'big':
83            snd_data.byteswap()
84        r.extend(snd_data)
85
86        silent = is_silent(snd_data)
87
88        if silent and snd_started:
89            num_silent += 1
90        elif not silent and not snd_started:
91            snd_started = True
92
93        if snd_started and num_silent > 30:
94            break
95
96    sample_width = p.get_sample_size(FORMAT)
97    stream.stop_stream()
98    stream.close()
99    p.terminate()
100
101    r = normalize(r)
102    r = trim(r)
103    r = add_silence(r, 0.5)
104    return sample_width, r
105
106def record_to_file(path):
107    "Records from the microphone and outputs the resulting data to 'path'"
108    sample_width, data = record()
109    data = pack('<' + ('h'*len(data)), *data)
110
111    wf = wave.open(path, 'wb')
112    wf.setnchannels(1)
113    wf.setsampwidth(sample_width)
114    wf.setframerate(RATE)
115    wf.writeframes(data)
116    wf.close()
117
118if __name__ == '__main__':
119    print("please speak a word into the microphone")
120    record_to_file('demo.wav')
121    print("done - result written to demo.wav")
122