Spaces:
Running
Running
| class ScoreBasis: | |
| def __init__(self, name=None): | |
| # the score operates on the specified rate | |
| self.score_rate = None | |
| # is the score intrusive or non-intrusive ? | |
| self.intrusive = True #require a reference | |
| self.name = name | |
| def windowed_scoring(self, audios, score_rate): | |
| raise NotImplementedError(f'In {self.name}, windowed_scoring is not yet implemented') | |
| def scoring(self, data, window=None, score_rate=None): | |
| """ calling the `windowed_scoring` function that should be specialised | |
| depending on the score.""" | |
| # imports | |
| #import soundfile as sf | |
| import resampy | |
| from museval.metrics import Framing | |
| #checking rate | |
| audios = data['audio'] | |
| score_rate = data['rate'] | |
| if self.score_rate is not None: | |
| score_rate = self.score_rate | |
| if score_rate != data['rate']: | |
| for index, audio in enumerate(audios): | |
| audio = resampy.resample(audio, data['rate'], score_rate, axis=0) | |
| audios[index] = audio | |
| if window is not None: | |
| framer = Framing(window * score_rate, window * score_rate, maxlen) | |
| nwin = framer.nwin | |
| result = {} | |
| for (t, win) in enumerate(framer): | |
| result_t = self.windowed_scoring([audio[win] for audio in audios], score_rate) | |
| result[t] = result_t | |
| else: | |
| result = self.windowed_scoring(audios, score_rate) | |
| return result | |
| """ | |
| audios = [] | |
| maxlen = 0 | |
| if isinstance(test_files, str): | |
| test_files = [test_files] | |
| print(f'test_files: {test_files}') | |
| if not self.intrusive and len(test_files) > 1: | |
| if self.verbose: | |
| print(' [%s] is non-intrusive. Processing first file only' | |
| % self.name) | |
| test_files = [test_files[0],] | |
| for file in test_files: | |
| # Loading sound file | |
| if isinstance(file, str): | |
| audio, rate = sf.read(file, always_2d=True) | |
| else: | |
| rate = array_rate | |
| if rate is None: | |
| raise ValueError('Sampling rate needs to be specified ' | |
| 'when feeding numpy arrays.') | |
| audio = file | |
| # Standardize shapes | |
| if len(audio.shape) == 1: | |
| audio = audio[:, None] | |
| if len(audio.shape) != 2: | |
| raise ValueError('Please provide 1D or 2D array, received ' | |
| '{}D array'.format(len(audio.shape))) | |
| if self.fixed_rate is not None and rate != self.fixed_rate: | |
| if self.verbose: | |
| print(' [%s] preferred is %dkHz rate. resampling' | |
| % (self.name, self.fixed_rate)) | |
| audio = resampy.resample(audio, rate, self.fixed_rate, axis=0) | |
| rate = self.fixed_rate | |
| if self.mono and audio.shape[1] > 1: | |
| if self.verbose: | |
| print(' [%s] only supports mono. Will use first channel' | |
| % self.name) | |
| audio = audio[..., 0, None] | |
| if self.mono: | |
| audio = audio[..., 0] | |
| maxlen = max(maxlen, audio.shape[0]) | |
| audios += [audio] | |
| audio = audios[1] | |
| audio[:maxlen-320] = audio[320:] | |
| audios[1] = audio | |
| for index, audio in enumerate(audios): | |
| if audio.shape[0] != maxlen: | |
| new = np.zeros((maxlen,) + audio.shape[1:]) | |
| new[:audio.shape[0]] = audio | |
| audios[index] = new | |
| if self.window is not None: | |
| framer = Framing(self.window * rate, | |
| self.hop * rate, maxlen) | |
| nwin = framer.nwin | |
| result = {} | |
| for (t, win) in enumerate(framer): | |
| result_t = self.test_window([audio[win] for audio in audios], | |
| rate) | |
| #or metric in result_t.keys(): | |
| # if metric not in result.keys(): | |
| # result[metric] = np.empty(nwin) | |
| # result[metric][t] = result_t[metric] | |
| result[t] = result_t | |
| else: | |
| result = self.test_window(audios, rate) | |
| return result | |
| """ | |