Source Code for Module timeside.analyzer.odf

  1  # -*- coding: utf-8 -*- 
  2  # 
  3  # Copyright (c) 2013 Paul Brossier <piem@piem.org> 
  4   
  5  # This file is part of TimeSide. 
  6   
  7  # TimeSide is free software: you can redistribute it and/or modify 
  8  # it under the terms of the GNU General Public License as published by 
  9  # the Free Software Foundation, either version 2 of the License, or 
 10  # (at your option) any later version. 
 11   
 12  # TimeSide is distributed in the hope that it will be useful, 
 13  # but WITHOUT ANY WARRANTY; without even the implied warranty of 
 14  # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the 
 15  # GNU General Public License for more details. 
 16   
 17  # You should have received a copy of the GNU General Public License 
 18  # along with TimeSide.  If not, see <http://www.gnu.org/licenses/>. 
 19   
 20  # Author: Thomas Fillon <thomas@parisson.com> 
 21   
 22  from timeside.core import implements, interfacedoc 
 23  from timeside.analyzer.core import Analyzer 
 24  from timeside.analyzer import Spectrogram 
 25  from timeside.api import IAnalyzer 
 26  import numpy as np 
 27  from numpy import pi as Pi 
 28  from scipy import signal 
 29   
 30   
 31 -class OnsetDetectionFunction(Analyzer): 
 32      implements(IAnalyzer) 
 33   
 34 -    def __init__(self, blocksize=1024, stepsize=None): 
 35          super(OnsetDetectionFunction, self).__init__() 
 36   
 37          self.input_blocksize = blocksize 
 38          if stepsize: 
 39              self.input_stepsize = stepsize 
 40          else: 
 41              self.input_stepsize = blocksize / 2 
 42   
 43          self.parents.append(Spectrogram(blocksize=self.input_blocksize, 
 44                              stepsize=self.input_stepsize)) 
 45   
 46      @interfacedoc 
 47 -    def setup(self, channels=None, samplerate=None, 
 48                blocksize=None, totalframes=None): 
 49          super(OnsetDetectionFunction, self).setup(channels, samplerate, 
 50                                                    blocksize, totalframes) 
 51   
 52      @staticmethod 
 53      @interfacedoc 
 54 -    def id(): 
 55          return "odf" 
 56   
 57      @staticmethod 
 58      @interfacedoc 
 59 -    def name(): 
 60          return "Onset Detection Function" 
 61   
 62      @staticmethod 
 63      @interfacedoc 
 64 -    def unit(): 
 65          return "" 
 66   
 67 -    def process(self, frames, eod=False): 
 68          return frames, eod 
 69   
 70 -    def post_process(self): 
 71   
 72          #spectrogram = self.parents()[0]['spectrogram_analyzer'].data 
 73          spectrogram = self.process_pipe.results['spectrogram_analyzer'].data 
 74          #spectrogram = self.pipe._results[self.parents()[0].id] 
 75   
 76          # Low-pass filtering of the spectrogram amplitude along the time axis 
 77          S = signal.lfilter(signal.hann(15)[8:], 1, abs(spectrogram), axis=0) 
 78   
 79   
 80          import matplotlib.pyplot as plt 
 81  #        plt.figure() 
 82  #        plt.imshow(np.log10(abs(spectrogram)), origin='lower', aspect='auto', interpolation='nearest') 
 83   
 84   
 85          # Clip small value to a minimal threshold 
 86          np.maximum(S, 1e-9, out=S) 
 87   
 88          S = np.log10(S) 
 89   
 90  #        plt.figure() 
 91  #        plt.imshow(S,origin='lower', aspect='auto', interpolation='nearest') 
 92  #        plt.show() 
 93   
 94          # S[S<1e-3]=0 
 95          np.maximum(S, 1e-3, out=S) 
 96   
 97          # Differentiator filter 
 98          df_filter = signal.fir_filter_design.remez(31, [0, 0.5], [Pi], 
 99                                                     type='differentiator') 
100   
101          S_diff = signal.lfilter(df_filter, 1, S, axis=0) 
102          S_diff[S_diff < 1e-10] = 0 
103   
104          # Summation along the frequency axis 
105          odf_diff = S_diff.sum(axis=1) 
106          odf_diff = odf_diff / np.median(odf_diff)  # Normalize 
107   
108          odf = self.new_result(data_mode='value', time_mode='framewise') 
109          #odf.parameters = {'FFT_SIZE': self.FFT_SIZE} 
110          odf.data_object.value = odf_diff 
111          self.process_pipe.results.add(odf) 
112