Update app.py

app.py

@@ -22,7 +22,8 @@ from pyannote.audio import Pipeline
 from pyannote.core import Annotation, Segment, Timeline
 import datetime as dt
 
-
+enableDenoise = False
+earlyCleanup = True
 
 # [None,Low,Medium,High,Debug]
 # [0,1,2,3,4]
@@ -81,14 +82,37 @@ def save_data(
     scheduler.append(data)
 
 def processFile(filePath):
+    global attenLimDb
+    global gainWindow
+    global minimumGain
+    global maximumGain
     print("Loading file")
-    waveform, sampleRate = torchaudio.load(filePath)
+    waveformList, sampleRate = su.splitIntoTimeSegments(filePath,600)
     print("File loaded")
+    enhancedWaveformList = []
+    if (enableDenoise):
+        print("Denoising")
+    for w in waveformList:
+        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")
-    diarization_output = pipeline({"waveform": waveform, "sample_rate": sampleRate})
+    diarization_output = pipeline({"waveform": waveform_gain_adjusted, "sample_rate": sampleRate})
     annotations = diarization_output.speaker_diarization
     print("Speakers Detected")
-    totalTimeInSeconds = int(waveform.shape[-1] / sampleRate)
+    totalTimeInSeconds = int(waveform_gain_adjusted.shape[-1]/sampleRate)
     print("Time in seconds calculated")
     return annotations, totalTimeInSeconds
     
@@ -311,6 +335,12 @@ sample_data = [f"CHEM1402_gt/24F_CHEM1402_Night_Class_Week_{i}_gt.rttm" for i in
 
 scheduler = ps.ParquetScheduler(repo_id="Sonogram/SampleDataset")
 
+secondDifference = 5
+gainWindow = 4
+minimumGain = -45
+maximumGain = -5
+attenLimDB = 3
+
 isGPU = False
 
 try:
@@ -326,6 +356,10 @@ except RuntimeError as e:
     print(f"Using {device} instead.")
 #device = xm.xla_device()
 
+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"))