Update app.py

app.py

@@ -23,6 +23,8 @@ from pyannote.core import Annotation, Segment, Timeline
 from df.enhance import enhance, init_df
 import datetime as dt
 
+enableDenoise = False
+earlyCleanup = True
 
 def save_data(
     config_dict: Dict[str,str], audio_paths: List[str], userid: str,
@@ -56,19 +58,28 @@ def processFile(filePath):
     waveformList, sampleRate = su.splitIntoTimeSegments(filePath,600)
     print("File loaded")
     enhancedWaveformList = []
-    print("Denoising")
+    if (enableDenoise):
+        print("Denoising")
     for w in waveformList:
-        newW = enhance(dfModel,dfState,w,atten_lim_db=attenLimDB).detach().cpu()
-        enhancedWaveformList.append(newW)
-    print("Audio denoised")
-    waveformEnhanced = su.combineWaveforms(enhancedWaveformList)     
+        if (enableDenoise):
+            newW = enhance(dfModel,dfState,w,atten_lim_db=attenLimDB).detach().cpu()
+            enhancedWaveformList.append(newW)
+        else:
+            enhancedWaveformList.append(w)
+    if (enableDenoise):
+        print("Audio denoised")
+    waveformEnhanced = su.combineWaveforms(enhancedWaveformList)    
+    if (earlyCleanup):
+        del enhancedWaveformList
     print("Equalizing Audio")
     waveform_gain_adjusted = su.equalizeVolume()(waveformEnhanced,sampleRate,gainWindow,minimumGain,maximumGain)
+    if (earlyCleanup):
+        del waveformEnhanced
     print("Audio Equalized")
     print("Detecting speakers")
-    annotations = pipeline({"waveform": waveformEnhanced, "sample_rate": sampleRate})
+    annotations = pipeline({"waveform": waveform_gain_adjusted, "sample_rate": sampleRate})
     print("Speakers Detected")
-    totalTimeInSeconds = int(waveformEnhanced.shape[-1]/sampleRate)
+    totalTimeInSeconds = int(waveform_gain_adjusted.shape[-1]/sampleRate)
     print("Time in seconds calculated")
     return annotations, totalTimeInSeconds
 
@@ -265,10 +276,10 @@ except RuntimeError as e:
     print(f"Using {device} instead.")
 #device = xm.xla_device()
 
-
-# Instantiate and prepare model for training.
-dfModel, dfState, _ = init_df(model_base_dir="DeepFilterNet3")
-dfModel.to(device)#torch.device("cuda"))
+if (enableDenoise):
+    # Instantiate and prepare model for training.
+    dfModel, dfState, _ = init_df(model_base_dir="DeepFilterNet3")
+    dfModel.to(device)#torch.device("cuda"))
 pipeline = Pipeline.from_pretrained("pyannote/speaker-diarization-3.1")
 pipeline.to(device)#torch.device("cuda"))