add_filter

main_pipeline.py

@@ -12,11 +12,20 @@ import pandas as pd
 import soundfile as sf
 
 
+def filter_small_speech(segments):
+    durs = segments.groupby('label').sum()
+    labels = durs[durs['duration'] / durs.sum()['duration'] > 0.015].index
+    return segments[segments.label.isin(labels)]
+
+
 def save_speaker_audios(segments, denoised_audio_path, out_folder='out', out_f=48000):
     signal, sr = librosa.load(denoised_audio_path, sr=out_f, mono=True)
     os.makedirs(out_folder, exist_ok=True)
     out_wav_paths = []
+
     segments = pd.DataFrame(segments)
+    segments['duration'] = segments.end - segments.start
+    segments = filter_small_speech(segments)
 
     for label in set(segments.label):
         temp_df = segments[segments.label == label]
@@ -45,7 +54,7 @@ def main_pipeline(audio_path):
 
 if __name__ == '__main__':
     parser = argparse.ArgumentParser()
-    parser.add_argument('--audio-path', default='dialog.mp3', help='Path to audio')
+    parser.add_argument('--audio-path', default='podkast.mp3', help='Path to audio')
     opt = parser.parse_args()
 
     for _ in tqdm(range(10)):