LocalAudioStream

[MikeiLL]

I'm working with P. Sobots ForeverFM module, which uses customized older version of Remix (and I think pyechonest).

Wondering if it would make sense to add the LocalAudioStream class to the audio module:

`class LocalAudioStream(AudioStream): """ Like a non-seekable LocalAudioFile with vastly better memory usage and performance. Takes a file-like object (and its kind, assumed to be MP3) and supports slicing and rendering. Attempting to read from a part of the input file that has already been read will throw an exception. """ def init(self, initializer, kind="mp3"): AudioStream.init(self, initializer)

    start = time.time()
    if hasattr(initializer, 'seek'):
        fobj = initializer
        fobj.seek(0)
    else:
        fobj = open(initializer, 'r')
    #   This looks like a lot of work, but is much more lighter
    #   on memory than reading the entire file in.
    md5 = hashlib.md5()
    while True:
        data = fobj.read(2 ^ 16)
        if not data:
            break
        md5.update(data)
    if not hasattr(initializer, 'seek'):
        fobj.close()
    track_md5 = md5.hexdigest()

    logging.getLogger(__name__).info("Fetching analysis...")
    try:
        tempanalysis = AudioAnalysis(str(track_md5))
    except EchoNestAPIError:
        tempanalysis = AudioAnalysis(initializer, kind)

    logging.getLogger(__name__).info("Fetched analysis in %ss",
                                     (time.time() - start))
    self.analysis = tempanalysis
    self.analysis.source = weakref.ref(self)

    class data(object):
        """
        Massive hack - certain operations are intrusive and check
        `.data.ndim`, so in this case, we fake it.
        """
        ndim = self.numChannels

    self.data = data

`

[tkell]

Ah, excellent, then you've already found the best way to deal with lots of audio.

We thought about adding Peter's streaming magic to the main repo, and decided against it - but for those people who want tons / streaming audio, it's a great place to start.