Functions | |
| computeFingerprint (x, f_s) | |
| computes subfingerprints from audio (derived from Haitsma et al.), 256 subfingerprints comprise one fingerprint | |
| generateBands_I (iFftLength, f_s) | |
| computeFingerprintCl (cPath) | |
| main | |
Variables | |
| parser = argparse.ArgumentParser(description='Extract fingerprint from wav file') | |
| metavar | |
| required | |
| help | |
| args = parser.parse_args() | |
| cPath = args.infile | |
Function Documentation
◆ computeFingerprint()
| computeFingerprint | ( | x, | |
| f_s ) |
computes subfingerprints from audio (derived from Haitsma et al.), 256 subfingerprints comprise one fingerprint
- Parameters
-
x array with floating point audio data (dimension samples x channels) f_s sample rate of audio data
- Returns
- F: series of subfingerprints
- t: time stamps
Definition at line 19 of file computeFingerprint.py.
19def computeFingerprint(x, f_s):
20
21 # set default parameters
22 fs_target = 5000
23 iBlockLength = 2048
24 iHopLength = 64
25
26 # pre-processing: down-mixing and normalization
27 x = ToolPreprocAudio(x)
28
29 # pre-processing: downsampling to target sample rate
30 if f_s != fs_target:
31 x, t_x = ToolResample(x, fs_target, f_s)
32
33 # initialization: generate transformation matrix for 33 frequency bands
34 H = generateBands_I(iBlockLength, fs_target)
35
36 # initialization: generate FFT window
37 afWindow = ToolComputeHann(iBlockLength)
38
39 # in the real world, we would do this block by block...
40 [X, f, tf] = computeSpectrogram(x, f_s, afWindow, iBlockLength, iHopLength)
41
42 # power spectrum
43 X = np.abs(X)**2
44
45 # group spectral bins in bands
46 E = np.matmul(H, X)
47
48 # extract fingerprint through diff (both time and freq)
49 SubFingerprint = np.diff(np.diff(E, 1, axis=0), 1, axis=1)
50 tf = tf[:-1] + iHopLength / (2 * fs_target)
51
52 # quantize fingerprint
53 SubFingerprint[SubFingerprint < 0] = 0
54 SubFingerprint[SubFingerprint > 0] = 1
55
56 return SubFingerprint, tf
57
58
Here is the call graph for this function:
◆ computeFingerprintCl()
| computeFingerprintCl | ( | cPath | ) |
main
Definition at line 83 of file computeFingerprint.py.
83def computeFingerprintCl(cPath):
84 from pyACA.ToolReadAudio import ToolReadAudio
85
86 # read audio file
87 [f_s, x] = ToolReadAudio(cPath)
88
89 # compute fingerprint
90 [F, t] = computeFingerprint(x, f_s)
91
92 return F, t
93
94
◆ generateBands_I()
| generateBands_I | ( | iFftLength, | |
| f_s ) |
Definition at line 59 of file computeFingerprint.py.
59def generateBands_I(iFftLength, f_s):
60
61 # constants
62 iNumBands = 33
63 f_max = 2000
64 f_min = 300
65
66 # initialize
67 f_band_bounds = f_min * np.exp(np.log(f_max / f_min) * range(iNumBands+1) / iNumBands)
68 f_fft = np.arange(iFftLength / 2 + 1) / iFftLength * f_s
69 H = np.zeros([iNumBands, iFftLength // 2 + 1])
70 idx = np.zeros([len(f_band_bounds), 2]).astype(int)
71
72 # get indices falling into each band
73 for k in range(len(f_band_bounds)-1):
74 idx[k, 0] = np.ceil(ToolFreq2Bin(f_band_bounds[k], iFftLength, f_s)).astype(int)
75 idx[k, 1] = np.floor(ToolFreq2Bin(f_band_bounds[k+1], iFftLength, f_s)).astype(int)
76 H[k, idx[k, 0]:idx[k, 1] + 1] = 1
77
78 return H
79
80
Here is the caller graph for this function:
Variable Documentation
◆ args
| args = parser.parse_args() |
Definition at line 104 of file computeFingerprint.py.
◆ cPath
| str cPath = args.infile |
Definition at line 105 of file computeFingerprint.py.
◆ help
| help |
Definition at line 101 of file computeFingerprint.py.
◆ metavar
| metavar |
Definition at line 100 of file computeFingerprint.py.
◆ parser
| parser = argparse.ArgumentParser(description='Extract fingerprint from wav file') |
Definition at line 99 of file computeFingerprint.py.
◆ required
| required |
Definition at line 100 of file computeFingerprint.py.
