source https://github.com/kaituoxu/Conv-TasNet

output_directory

which contain the mix audio and the separate audio and use pretrained model final.pth.tar to separate the mix audio

Changes on code

Changes in EvalDataset Class

File: src/data.py

Before:

class EvalDataset(data.Dataset):
    def __init__(self, mix_dir, mix_json, batch_size, sample_rate=8000):
        ...

After:

class EvalDataset(data.Dataset):
    def __init__(self, mix_dir, mix_json, batch_size, sample_rate=8000):
        ...
        with open(mix_json, 'r') as f:
            mix_infos = json.load(f)['mix']
        ...

Changes in load_mixtures_and_sources Function

File: src/data.py

Before:

def load_mixtures_and_sources(batch):
    ...
    for mix_info, s1_info, s2_info in zip(mix_infos, s1_infos, s2_infos):
        mix_path = mix_info[0]
        s1_path = s1_info[0]
        s2_path = s2_info[0]
        ...

After:

def load_mixtures_and_sources(batch):
    ...
    for mix_info, s1_info, s2_info in zip(mix_infos, s1_infos, s2_infos):
        mix_path = mix_info['mix_path']
        s1_path = s1_info['mix_path']
        s2_path = s2_info['mix_path']
        ...

Changes in load_mixtures Function

File: src/data.py

Before:

def load_mixtures(batch):
    ...
    for mix_info in mix_infos:
        mix_path = mix_info[0]
        ...

After:

def load_mixtures(batch):
    ...
    for mix_info in mix_infos:
        mix_path = mix_info['mix_path']
        ...

Adding and Modifying write Function

File: src/separate.py

Before:

def write(inputs, filename, sr):
    librosa.output.write_wav(filename, inputs, sr)

After:

import soundfile as sf

def write(inputs, filename, sr):
    sf.write(filename, inputs, sr)

Changes in separate Function

File: src/separate.py

Before:

def separate(args):
    ...
    eval_loader = EvalDataLoader(eval_dataset, batch_size=1)
    ...
    def write(inputs, filename, sr):
        librosa.output.write_wav(filename, inputs, sr)
    ...
            write(mixture[i], filename + '.wav')
            for c in range(C):
                write(flat_estimate[i][c], filename + '_s{}.wav'.format(c+1))

After:

def separate(args):
    ...
    eval_loader = DataLoader(eval_dataset, batch_size=1, shuffle=False, collate_fn=_collate_fn_eval)
    ...
    def write(inputs, filename, sr):
        sf.write(filename, inputs, sr)
    ...
            write(mixture[i].cpu().numpy(), filename + '.wav', args.sample_rate)
            for c in range(C):
                write(flat_estimate[i][c].cpu().numpy(), filename + '_s{}.wav'.format(c+1), args.sample_rate)