Getting max amplitude for an audio file per second

Question

I know there are some similar questions here, but most of them are concerning generating waveform images, which is not what I want.

My goal is to generate a waveform visualization for an audio file, similar to SoundCloud, but not an image. I'd like to have the max amplitude data for each second (or half second) of an audio clip in an array. I could then use this data to create a CSS-based visualization.

Ideally I'd like to get an array that has all the amplitude values for each second as a percentage of the maximum amplitude of the entire audio file. Here's an example:

[
    0.0,  # Relative max amplitude of first second of audio clip (0%)
    0.04,  # Relative max amplitude of second second of audio clip (4%)
    0.15,  # Relative max amplitude of third second of audio clip (15%)
    # Some more
    1.0,  # The highest amplitude of the whole audio clip will be 1.0 (100%)
]

I assume I'll have to use at least numpy and Python's wave module, but I'm not sure how to get the data I want. I'd like to use Python but I'm not completely against using some kind of command-line tool.

Answer 1

If you allow gstreamer, here is a little script that could do the trick. It accept any audio file that gstreamer can handle.

• Construct a gstreamer pipeline, use audioconvert to reduce the channels to 1, and use level module to get peaks
• Run the pipeline until EOS is hit
• Normalize the peaks from the min/max found.

Snippet:

import os, sys, pygst
pygst.require('0.10')
import gst, gobject
gobject.threads_init()

def get_peaks(filename):
    global do_run

    pipeline_txt = (
        'filesrc location="%s" ! decodebin ! audioconvert ! '
        'audio/x-raw-int,channels=1,rate=44100,endianness=1234,'
        'width=32,depth=32,signed=(bool)True !'
        'level name=level interval=1000000000 !'
        'fakesink' % filename)
    pipeline = gst.parse_launch(pipeline_txt)

    level = pipeline.get_by_name('level')
    bus = pipeline.get_bus()
    bus.add_signal_watch()

    peaks = []
    do_run = True

    def show_peak(bus, message):
        global do_run
        if message.type == gst.MESSAGE_EOS:
            pipeline.set_state(gst.STATE_NULL)
            do_run = False
            return
        # filter only on level messages
        if message.src is not level or \
           not message.structure.has_key('peak'):
            return
        peaks.append(message.structure['peak'][0])

    # connect the callback
    bus.connect('message', show_peak)

    # run the pipeline until we got eos
    pipeline.set_state(gst.STATE_PLAYING)
    ctx = gobject.gobject.main_context_default()
    while ctx and do_run:
        ctx.iteration()

    return peaks

def normalize(peaks):
    _min = min(peaks)
    _max = max(peaks)
    d = _max - _min
    return [(x - _min) / d for x in peaks]

if __name__ == '__main__':
    filename = os.path.realpath(sys.argv[1])
    peaks = get_peaks(filename)

    print 'Sample is %d seconds' % len(peaks)
    print 'Minimum is', min(peaks)
    print 'Maximum is', max(peaks)

    peaks = normalize(peaks)
    print peaks

And one output example:

$ python gstreamerpeak.py 01\ Tron\ Legacy\ Track\ 1.mp3 
Sample is 182 seconds
Minimum is -349.999999922
Maximum is -2.10678956719
[0.0, 0.0, 0.9274581631597019, 0.9528318436488018, 0.9492396611762614,
0.9523404330322813, 0.9471685835966183, 0.9537281219301242, 0.9473486577135167,
0.9479292126411365, 0.9538221105563514, 0.9483845795252251, 0.9536790832823281,
0.9477264933378022, 0.9480077366961968, ...