Update inference/style_transfer.py

inference/style_transfer.py

@@ -21,6 +21,7 @@ currentdir = os.path.dirname(os.path.realpath(__file__))
 sys.path.append(os.path.join(os.path.dirname(currentdir), "mixing_style_transfer"))
 from networks import FXencoder, TCNModel
 from data_loader import *
+import librosa
 
 
 
@@ -304,7 +305,23 @@ class Mixing_Style_Transfer_Inference:
         return whole_batch_data
 
 
-def set_up():
+
+def trim_audio(target_file_path, start_point_in_second=0, duration_in_second=30, sample_rate=44100):
+    # insure format
+    cur_aud, _ = librosa.load(target_file_path, sr=sample_rate, mono=False)
+    sf.write(target_file_path, cur_aud.transpose(-1, -2), sample_rate, 'PCM_16')
+    # trim if possible
+    cur_wav_length = load_wav_length(target_file_path)
+    if cur_wav_length < duration_in_second*sample_rate:
+        return
+    if cur_wav_length-start_point_in_second*sample_rate < duration_in_second*sample_rate:
+        trimmed_audio = load_wav_segment(target_file_path, start_point=start_point_in_second*sample_rate, axis=1)
+    else:
+        trimmed_audio = load_wav_segment(target_file_path, start_point=start_point_in_second*sample_rate, duration=duration_in_second*sample_rate, axis=1)
+    sf.write(target_file_path, trimmed_audio, sample_rate, 'PCM_16')
+
+
+def set_up(start_point_in_second=0, duration_in_second=30):
     os.environ['MASTER_ADDR'] = '127.0.0.1'
     os.environ["CUDA_VISIBLE_DEVICES"] = '0'
     os.environ['MASTER_PORT'] = '8888'