#1 Dataloder 구성 관련

dahyunnss commented 1 year ago

code issue

def __getitem__(self, index): #issue
        # first, get target track
        index = index // self.samples_per_track
        print(3333)
        print(self.tracks)
        track_path = self.tracks[index]['path']
        min_duration = self.tracks[index]['min_duration']
        if self.random_chunks:
            start = random.randint(0, min_duration - self.seq_duration)
        else:
            start = 0

        audio_sources = []

        target_audio = load_audio(
            track_path / self.target_file, start=start, dur=self.seq_duration
        )
        target_audio = self.source_augmentations(target_audio)
        audio_sources.append(target_audio)

        for source in self.interferer_files:
            if self.random_track_mix:
                random_idx = random.choice(range(len(self.tracks)))
                track_path = self.tracks[random_idx]['path']
                if self.random_chunks:
                    min_duration = self.tracks[random_idx]['min_duration']
                    start = random.randint(0, min_duration - self.seq_duration)

            audio = load_audio(
                track_path / source, start=start, dur=self.seq_duration
            )

            audio = self.source_augmentations(audio)
            audio_sources.append(audio)

        stems = torch.stack(audio_sources)
        # # apply linear mix over source index=0
        x = stems.sum(0)

        # y = stems.reshape(-1, stems.size(2))
        y = stems[0]

        return x, y

def get_tracks(self):
        p = Path(self.root, self.split)
        print(p)
        for track_path in tqdm.tqdm(p.iterdir(), disable=True):
            print('ss', track_path.is_dir())
            if track_path.is_dir():
                source_paths = [track_path / s for s in self.source_files]
                print('pp',source_paths)
                if not all(sp.exists() for sp in source_paths):
                    print("exclude track ", track_path)
                    continue

                if self.seq_duration is not None:
                    infos = list(map(load_info, source_paths))
                    # get minimum duration of track
                    min_duration = min(i['duration'] for i in infos)
                    if min_duration > self.seq_duration:
                        yield ({
                            'path': track_path,
                            'min_duration': min_duration
                        })
                else:
                    yield ({'path': track_path, 'min_duration': None})

Error log Traceback (most recent call last): File "train.py", line 214, in main() File "train.py", line 127, in main item = train_dataset[0] File "/userHome/userhome1/sojeong/voice/Singing_Voice_Synthesis/U-net/data.py", line 93, in getitem track_path = self.tracks[index]['path'] IndexError: list index out of range

feVeRin commented 1 year ago

내 생각에 바로 loader 구성하는 것보다 라이브러리 거쳐서 하는게 더 편할 것 같아서
dataloader.ipynb 파일에 musdb 라이브러리 사용 관련해서 code 업데이트 했음
해당 파일에서 musdb 라이브러리 통해서 vocal을 label로 두고 dataset의 오디오를 처리해보면 (2,220500) shape로 나옴
트랙 속성도 정리 해놓았으니까 라이브러리 쓸 때 참고하면 좋을 듯

feVeRin commented 1 year ago

dataloader 수정 완료 https://github.com/feVeRin/Singing_Voice_Synthesis/commit/5ac5cce2510072129e28279806575ac00af73d35

feVeRin / Singing_Voice_Synthesis

#1 Dataloder 구성 관련 #1

code issue