add_filter

main_pipeline.py

@@ -13,6 +13,7 @@ import soundfile as sf
 
 
 def filter_small_speech(segments):
+    segments['duration'] = segments.end - segments.start
     durs = segments.groupby('label').sum()
     labels = durs[durs['duration'] / durs.sum()['duration'] > 0.015].index
     return segments[segments.label.isin(labels)]
@@ -24,7 +25,6 @@ def save_speaker_audios(segments, denoised_audio_path, out_folder='out', out_f=4
     out_wav_paths = []
 
     segments = pd.DataFrame(segments)
-    segments['duration'] = segments.end - segments.start
     segments = filter_small_speech(segments)
 
     for label in set(segments.label):
@@ -54,7 +54,7 @@ def main_pipeline(audio_path):
 
 if __name__ == '__main__':
     parser = argparse.ArgumentParser()
-    parser.add_argument('--audio-path', default='podkast.mp3', help='Path to audio')
+    parser.add_argument('--audio-path', default='dialog.mp3', help='Path to audio')
     opt = parser.parse_args()
 
     for _ in tqdm(range(10)):

utils/denoise_pipeline.py

@@ -1,10 +1,8 @@
-import os
-import torch
-from scipy.io.wavfile import write
 import librosa
 import torch
 from demucs.apply import apply_model
 from demucs.pretrained import get_model
+from scipy.io.wavfile import write
 
 # demucs_model = get_model('cfa93e08')
 demucs_model = get_model('htdemucs')