Python wavfile.read函数代码示例

    def test_realFile(self):
        original_file = self.auxiliary_files_url + "/nai_sample.wav"
        denoised_file = self.auxiliary_files_url + "/nai_sample_sox_denoised.wav"
        user = 'rapp'
        audio_type = 'nao_wav_1_ch'
        scale = 0.2

        result = self.sox_denoise_module.soxDenoise(\
                user,\
                audio_type,\
                original_file,\
                denoised_file,\
                scale)
        # The function thinks the denoising succeded
        self.assertEqual(result, "true")
        # Check for the denoised file
        denoised_exists = os.path.isfile(denoised_file)
        self.assertEqual(denoised_exists, True)
        # Check if denoised energy is lower than the initial one
        samp_freq, signal_orig = wavfile.read(original_file)
        energy_orig = 0.0
        for i in signal_orig:
            energy_orig += i * 1.0 * i
        samp_freq, signal_denoised = wavfile.read(denoised_file)
        energy_denoised = 0.0
        for i in signal_denoised:
            energy_denoised += i * 1.0 * i
        self.assertGreater(energy_orig, energy_denoised)

        # erase the denoised file
        os.remove(denoised_file)

def generate_mixture(src1, src2, fname, attn1, attn2):
    """
        mixes 10 seconds of two sources of the same sample rate and saves them as fname

    Args:
        src1: filename for the first source
        src2: filename for the second source
        fname: output filename to save as
        attn1: relative attenuation for the first source
        attn2: relative attenuation for the second source

    Returns:

    """
    sr1, data1 = wav.read(src1)
    if data1.dtype == np.dtype("int16"):
        data1 = data1 / float(np.iinfo(data1.dtype).max)

    sr2, data2 = wav.read(src2)
    if data2.dtype == np.dtype("int16"):
        data2 = data2 / float(np.iinfo(data2.dtype).max)

    if sr1 != sr2:
        raise ValueError("Both sources muse have same sample rate")

    attn1 = float(attn1 + 1) / 2
    attn2 = float(attn2 + 1) / 2
    sample1 = data1[0:10 * sr1]
    sample2 = data2[0:10 * sr1]
    left = attenuate(sample1, attn1) + attenuate(sample2, attn2)
    right = attenuate(sample1, 1-attn1) + attenuate(sample2, 1-attn2)

    signal = np.vstack((left, right))
    scipy.io.wavfile.write(fname, sr1, signal.T)

def generate_reverb(signal, reverb, fname, iter_range):
    """
    Adds reverb from the path reverb to the data in the path signal and saves it as fname. Applies reverb iteratively over
    iter_range
    :param signal: the filename for the stereo input signal
    :param reverb: the filename for the stereo impulse response
    :param fname: the output filename to save as
    :param iter_range: the max number of iterations to convolve with the signal
    :return:
    """
    sr, data = wav.read(signal)
    if data.dtype == np.dtype("int16"):
        data = data / float(np.iinfo(data.dtype).max)


    sr_ir, data_ir = wav.read(reverb)
    if data_ir.dtype == np.dtype("int16"):
        data_ir = data_ir / float(np.iinfo(data_ir.dtype).max)

    if sr_ir != sr:
        raise ValueError("Impulse Response must have same sample rate as signal")

    prev_data = data
    for i in xrange(0, iter_range+1):
        if i > 0:
            mix = add_reverb(prev_data.T, data_ir.T)
            prev_data = np.copy(mix).T
        else:
            mix = data.T
        if not os.path.exists(os.path.splitext(fname)[0]+'-'+str(i)+'.wav'):
            scipy.io.wavfile.write(os.path.splitext(fname)[0]+'-'+str(i)+'.wav', sr, mix.T)

def load_data(syllable, N, used_samples, snr, sample_order = None):
    """Function that goes through all N samples of syllable and loads its wave data.
    
    :param syllable: complete path name of syllable (string)
    :param N: number of samples to load
    :param used_samples: number of samples to skip in the beginning
    :param snr: the strength of the noise
    :param sample_order: if not None should be vector of indices of samples to be loaded (default = None)
    
    :returns syllable_waves: list of N sample waves of syllable
    """

    samples = [files for files in os.listdir(syllable)]
    syllable_waves = []
    if sample_order is None:
        for i in range(int(N)):
            rate, wave = wav.read(syllable + '/' + samples[i + used_samples])
            if (snr != 0.0):
                noiseLvl = np.sqrt(np.var(wave) / snr)
            else:
                noiseLvl = 0.0
            wave = wave + noiseLvl * np.random.randn(len(wave))
            syllable_waves.append([wave,rate])
    else:
        for i in sample_order:
            rate, wave = wav.read(syllable + '/' + samples[i])
            if(snr != 0.0):
                noiseLvl = np.sqrt(np.var(wave) / snr)
            else:
                noiseLvl = 0.0
            wave = wave + noiseLvl * np.random.randn(len(wave))
            syllable_waves.append([wave,rate])
    return syllable_waves

def simple_noise_filter(target, files, method=median_by_intensity, combination=flatten, section_length=4096):
    # load all .mp3 files into an arrays 
    # bin each to a certain length
    #print time()
    feeds = [section_by_length(wavfile.read(file)[1], section_length) for file in files]
    samplerate = wavfile.read(files[0])[0]
    #print time()
    # perform fft on each bin, select median of each 
    max_len = len(max(feeds, key=len))
    sections = []
    for i in range(max_len):
        begin = time()
        freqs = [fft.fft(feed[i], axis=0) for feed in feeds]
        #print "Fourier per ~.1s feed: ",
        #print (time()-begin)/3.
        begin = time()
        #filtered_freqs = [median_by_intensity(freqs, j) for j in range(len(freqs[0]))] # traverse the arrays in parallel
        filtered_freqs = [method(freqs, j) for j in range(len(freqs[0]))]
        #print "Filtering: ",
        #print (time()-begin)
        begin = time()
        sections += [real(fft.ifft(filtered_freqs, axis=0)).astype(feeds[0][0].dtype)]
        #print "Inversing per ~.1s feed: ",
        #print (time() - begin)
    # output
    #print time()
    samples = combination(sections)
    wavfile.write(target, samplerate, samples)

def test_ubm_var_channel():
    ubm = GMM.load('model/ubm.mixture-32.person-20.immature.model')

    train_duration = 8.
    nr_test = 5
    test_duration = 3.
    audio_files = ['xinyu.vad.wav', 'wyx.wav']
    X_train, y_train, X_test, y_test = [], [], [], []
    for audio_file in audio_files:
        fs, signal = wavfile.read(audio_file)
        signal = monotize_signal(signal)

        train_len = int(fs * train_duration)
        test_len = int(fs * test_duration)

        X_train.append(mix_feature((fs, signal[:train_len])))
        y_train.append(audio_file)

        for i in range(nr_test):
            start = random.randint(train_len, len(signal) - test_len)
            X_test.append(mix_feature((fs, signal[start:start+train_len])))
            y_test.append(audio_file)

    gmmset = GMMSet(32, ubm=ubm)
    gmmset.fit(X_train, y_train)
    y_pred = gmmset.predict_with_reject(X_test)
    for i in xrange(len(y_pred)):
        print y_test[i], y_pred[i], '' if y_test[i] == y_pred[i] else 'wrong'

    for imposter_audio_file in map(
            lambda x: 'test-{}.wav'.format(x), range(5)):
        fs, signal = wavfile.read(imposter_audio_file)
        signal = monotize_signal(signal)
        imposter_x = mix_feature((fs, signal))
        print gmmset.predict_one_with_rejection(imposter_x)

def estim_diff(percent=256):
    sound_counter=0
    res=np.empty(len(input_file_names))
    for i in range(res.shape[0]):
        input_rate,input_sig=wavfile.read(input_dir+'Segments/'+input_file_names[i])
        output_rate,output_sig=wavfile.read(output_dir+'Segments/'+output_file_names[i])
    
        input_sig=pcm2float(input_sig,'float32')
        output_sig=pcm2float(output_sig,'float32')
        
        min_size=np.min((input_sig[:,0].shape[0],output_sig[:,0].shape[0]))
        #print min_size,min_size*percent
        #S_inp=np.absolute(fft(input_sig[:min_size,0]-np.mean(input_sig[:min_size,0])))
        #S_out=np.absolute(fft(output_sig[:min_size,0]-np.mean(output_sig[:min_size,0])))
    
        t=time()
        nperseg=int(min_size*percent)-np.mod(int(min_size*percent),10)
        real_perc=float(float(nperseg)/int(min_size*percent))
        S_inp=signal.welch(input_sig[:min_size,0],nperseg=nperseg)[1]    
        S_out=signal.welch(output_sig[:min_size,0],nperseg=nperseg)[1]    
        #S_inp=ndim_welch(input_sig[:min_size,0][None,...],nperseg=int(min_size*percent))[1]    
        #S_out=ndim_welch(output_sig[:min_size,0][None,...],nperseg=int(min_size*percent))[1]    
        #print time()-t
        #print S_inp_1,S_inp_2
        res[sound_counter]=delta_estimator_3(S_out/S_inp,S_inp)-delta_estimator_3(S_inp/S_out,S_out)
        #out=float2pcm(output_sig,'int16')
        sound_counter+=1
    return real_perc,int(min_size*percent),res

def find_offset(file1, file2, fs=8000, trim=60*15, correl_nframes=1000):
    tmp1 = convert_and_trim(file1, fs, trim)
    tmp2 = convert_and_trim(file2, fs, trim)
    # Removing warnings because of 18 bits block size
    # outputted by ffmpeg
    # https://trac.ffmpeg.org/ticket/1843
    warnings.simplefilter("ignore", wavfile.WavFileWarning)
    a1 = wavfile.read(tmp1, mmap=True)[1] / (2.0 ** 15)
    a2 = wavfile.read(tmp2, mmap=True)[1] / (2.0 ** 15)
    # We truncate zeroes off the beginning of each signals
    # (only seems to happen in ffmpeg, not in sox)
    a1 = ensure_non_zero(a1)
    a2 = ensure_non_zero(a2)
    mfcc1 = mfcc(a1, nwin=256, nfft=512, fs=fs, nceps=13)[0]
    mfcc2 = mfcc(a2, nwin=256, nfft=512, fs=fs, nceps=13)[0]
    mfcc1 = std_mfcc(mfcc1)
    mfcc2 = std_mfcc(mfcc2)
    c = cross_correlation(mfcc1, mfcc2, nframes=correl_nframes)
    max_k_index = np.argmax(c)
    # The MFCC window overlap is hardcoded in scikits.talkbox
    offset = max_k_index * 160.0 / float(fs) # * over / sample rate
    score = (c[max_k_index] - np.mean(c)) / np.std(c) # standard score of peak
    os.remove(tmp1)
    os.remove(tmp2)
    return offset, score

def perf_eval(param):
    # wrtitten in 3000 basis, finding the nperseg value from param
    nperseg=param % 3000
    # wrtitten in 3000 basis, finding the number of music segments
    num_of_seg_idx=(param-nperseg)/3000
    num_of_seg=num_of_segs[num_of_seg_idx]

    input_rate,input_sig=wavfile.read(input_dir+song_name+'.wav')
    output_rate,output_sig=wavfile.read(output_dir+song_name+'.wav')
    
    #the +1 in denominator is because we exclude the last piece of music to only
    # consider music pieces of the same size.
    input_seg_len=input_sig.shape[0]/(num_of_seg+1)
    output_seg_len=output_sig.shape[0]/(num_of_seg+1)
    
    if input_rate!=output_rate:
        print ("Rate Mistmatch!")
        sys.exit(0)
    
    #print (nperseg,nperseg_step,input_seg_len,output_seg_len)
    if np.min((input_seg_len,output_seg_len))*0.7 < nperseg * nperseg_step:
        print ("Nothing to do!")
        sys.exit(0)
        
    res=estim_diff(input_sig, input_seg_len, output_sig, output_seg_len, nperseg, num_of_seg, nperseg_step)

    
    f=open('/agbs/cluster/naji/Linear Filters/Echo/out/Winter/Room/'+str(num_of_seg)+'/'+str(nperseg)+'.txt','w')
    print (nperseg,file=f)
    print (np.mean(res>0),file=f)

def get_offset_wav(wav_filename1, wav_filename2, time_limit=300):
    """Return offset in seconds between wav_filename1 and
    wav_filename2, which are recordings of the same event
    with potentially different starting times. Returns the 
    number of seconds that wav_filename2 starts after wav_filename1
    (possibly negative).
     
    
    If time_limit is provided, clip files
    to first time_limit seconds. This can substantially speed up 
    offset detection"""
    
    rate1, data1 = sp_wav.read(wav_filename1)
    rate2, data2 = sp_wav.read(wav_filename2)
    # the two files must have the same sampling rate
    assert(rate1==rate2)
    
    if time_limit is not None:
        data1 = data1[0:rate1 * time_limit]
        data2 = data2[0:rate2 * time_limit]
                
    offset_samples = get_offset_xcorr(data1, data2)
    offset_seconds = offset_samples / float(rate1)
    
    return offset_seconds

def mix_files(f1,f2):

    base1 = f1.split('/')[-1].split('.wav')[0]
    base2 = f2.split('/')[-1].split('.wav')[0]

    (fs,sig) = wav.read(f1)
    s1 = sig.reshape((len(sig),1))
    del sig
    (fs,sig) = wav.read(f2)
    s2 = sig.reshape((len(sig),1))
    del sig
    block_length = 5*fs
    s1_blocks = enframe(s1,block_length,block_length)
    s2_blocks = enframe(s2,block_length,block_length)
    del s1, s2
     
    nrg1 = 0.707*np.sqrt(np.sum(np.power(s1_blocks,2),axis=1))
    nrg2 = 0.707*np.sqrt(np.sum(np.power(s2_blocks,2),axis=1))
     
    for i in range(len(nrg1)):
        db1 = np.log(nrg1[i])
        db2 = np.log(nrg2[i])
        if (db1 >= 9) and (db2 >= 9) and (0.1 < abs(db1 - db2) < 5):
            sir = '%.2f' % (db1 - db2)
            ovl_name = '/erasable/nxs113020/wav_ovl/'+base1+'_'+base2+'_sir'+sir+'_'+str(i)+'.wav'
            overlapped = s1_blocks[i,:] + s2_blocks[i,:]
            nrg_ovl = 0.707*np.sqrt(np.sum(np.power(overlapped,2)))
            scikits.audiolab.wavwrite(overlapped/nrg_ovl, ovl_name, fs, 'pcm16')

    def test_write_edge_values(self):
        # Write edge values 1.0
        samples = numpy.ones((441, 1), dtype=numpy.float32)
        dest_file = NamedTemporaryFile(delete=True)
        wfile, infos = wav.open_write_mode(dest_file.name, 44100, 1)
        wav.write_block(wfile, samples)
        wfile._file.flush() # To force the file to be written to the disk

        frame_rate, samples_written = sp_wavfile.read(dest_file.name)
        numpy.testing.assert_array_equal(samples_written, numpy.array([2**15 - 1] * 441, dtype=numpy.int16))
        dest_file.close()

        # Write value 2.0, clipped to 1.0
        samples = numpy.ones((441, 1), dtype=numpy.float32) * 2.0
        dest_file = NamedTemporaryFile(delete=True)
        wfile, infos = wav.open_write_mode(dest_file.name, 44100, 1)
        wav.write_block(wfile, samples)
        wfile._file.flush() # To force the file to be written to the disk

        frame_rate, samples_written = sp_wavfile.read(dest_file.name)
        numpy.testing.assert_array_equal(samples_written, numpy.array([2**15 - 1] * 441, dtype=numpy.int16))
        dest_file.close()

        # Write edge values -1.0
        samples = numpy.ones((441, 1), dtype=numpy.float32) * -1
        dest_file = NamedTemporaryFile(delete=True)
        wfile, infos = wav.open_write_mode(dest_file.name, 44100, 1)
        wav.write_block(wfile, samples)
        wfile._file.flush() # To force the file to be written to the disk

        frame_rate, samples_written = sp_wavfile.read(dest_file.name)
        numpy.testing.assert_array_equal(samples_written, numpy.array([-2**15] * 441, dtype=numpy.int16))
        dest_file.close()

def read_sound(fp):
    """
    create a normalized float array and datarate from any audo file
    """
    if fp.endswith('mp3'):
        try:
            oname = 'temp.wav'
            #cmd = 'lame --decode "{0}" {1}'.format( fp ,oname )
            result = subprocess.call(['lame', '--decode', fp, oname])
            assert(result is 0)
            samplerate, data = wav.read(oname)
        except:
            print "couldn't run lame"
            try:
                import moviepy.editor as mpy
                aud_clip = mpy.AudioFileClip(fp)
                samplerate = aud_clip.fps
                data = aud_clip.to_soundarray()
            except:
                print "moviepy not installed?"
    if fp.endswith('aif'):
        #sf = aifc.open(fp)
        oname = fp
        sf = Sndfile(fp, 'r')
        sf.seek(0)
        data = sf.read_frames(sf.nframes)
        samplerate = sf.samplerate
    if fp.endswith('wav'):
        samplerate, data = wav.read(fp)

    if len(data.shape)>1: data = data[:,0]
    data = data.astype('float64')
    data /= data.max()
    return data, samplerate

def plot_from_wavfile(file1, file2):
    '''
    Given two wav files, plot their frequency spectrums
    '''
    rate1, data1 = wavefile.read(file1)
    rate2, data2 = wavefile.read(file2)
    plot_from_rawdata(data1, data2, rate1)

def processing():
    """post-processing of MLSbuf and recBuf, using the matched filter functions"""
    # -- start recording and playback in async. mode
    play_while_recording()
    
    global SAMPLE_RATE

    # -- latency for input and output devs, obtained using portaudio pa_devs script
    inputLatency = 0.0087
    outputLatency = 0.0087
    # -- convert latencies to num. of samples
    latencySamples = math.ceil((inputLatency+outputLatency)*SAMPLE_RATE)

    # -- calibration samples (uncomment for debugging)
    calSamp = 52
    
    # -- load recording buffer into numpy array
    recData = read("recBuf.wav")
    recBuf = np.array(recData[1],dtype =float)
    # -- index of internal delays & calibritation samples to subtract
    interDelaySamp = np.s_[0:(latencySamples + calSamp)]
    recBuf = np.delete(recBuf,interDelaySamp)
    # -- remove excess samples from the recording buffer
    removeExcessSamples = np.s_[6000:]
    recBuf = np.delete(recBuf,removeExcessSamples)
    
    # -- load playback buffer
    MLSdata = read("MLS.wav")
    MLSbuf = np.array(MLSdata[1],dtype =float)
 
    # -- compute delay using Matched Filters & normalize
    xcorr = matched_filter(MLSbuf,recBuf)/50000000000.0

    # -- get gain
    gain = get_gain(MLSbuf,recBuf)
    
    # -- peak detector
    prop_delay = peak_detector(xcorr)

    # -- plot recorded seq, Tx MLS seq. (uncomment for debugging)
    plt.figure(1)
    plt.plot(MLSbuf)
    plt.title("MLS sequence")
    plt.xlabel("samples")
    plt.grid(True)
    
    plt.figure(2)
    plt.plot(recBuf)
    plt.title("Recorded MLS sequence")
    plt.xlabel("samples")
    plt.ylabel("Amplitude")
    plt.grid(True)

    plt.figure(3)
    plt.plot(abs(xcorr))
    plt.title("Matched Filter Output")
    plt.xlabel("delay (samples)")
    plt.ylabel("Rxy")
    plt.grid(True)
    plt.show()