Update app.py

app.py

@@ -16,6 +16,10 @@ import pandas as pd
 import torchaudio
 
 
+from torchmetrics.audio import ShortTimeObjectiveIntelligibility as STOI
+from torchmetrics.audio.pesq import PerceptualEvaluationSpeechQuality as PESQ
+
+
 
 @st.cache
 def load_model():
@@ -128,7 +132,52 @@ if st.button('Сгенерировать потери'):
     st.text('Улучшенное аудио')
     st.audio('enhanced.wav')
 
-    
+
+
+
+
+
+    data_clean, samplerate = torchaudio.load('/content/Катя_базу_выдала.wav')
+    data_lossy, samplerate = torchaudio.load('/content/Катя_базу_выдала_40%.wav')
+    data_enhanced, samplerate = torchaudio.load('/content/Катя_базу_выдала_демо.wav')
+
+    min_len = min(data_clean.shape[1], data_lossy.shape[1], data_enhanced.shape[1])
+    data_clean = data_clean[:, :min_len]
+    data_lossy = data_lossy[:, :min_len]
+    data_enhanced = data_enhanced[:, :min_len]
+
+
+    stoi = STOI(samplerate)
+
+    stoi_orig = round(float(stoi(data_clean, data_clean)),3)
+    stoi_lossy = round(float(stoi(data_clean, data_lossy)),5)
+    stoi_enhanced = round(float(stoi(data_clean, data_enhanced)),5)
+
+    stoi_mass=[stoi_orig, stoi_lossy, stoi_enhanced]
+
+
+    pesq = PESQ(16000, 'nb')
+
+    data_clean = data_clean.cpu().numpy()
+    data_lossy = data_lossy.cpu().numpy()
+    data_enhanced = data_enhanced.cpu().numpy()
+
+    if samplerate != 16000:
+        data_lossy = librosa.resample(data_lossy, orig_sr=48000, target_sr=16000)
+        data_clean = librosa.resample(data_clean, orig_sr=48000, target_sr=16000)
+        data_enhanced = librosa.resample(data_enhanced, orig_sr=48000, target_sr=16000)
+
+    pesq_orig = float(pesq(torch.tensor(data_clean), torch.tensor(data_clean)))
+    pesq_lossy = float(pesq(torch.tensor(data_lossy), torch.tensor(data_clean)))
+    pesq_enhanced = float(pesq(torch.tensor(data_enhanced), torch.tensor(data_clean)))
+
+    psq_mas=[pesq_orig, pesq_lossy, pesq_enhanced]
+
+
+
+
+
+    #_____________________________________________
     data_clean, samplerate = sf.read('target.wav')
     data_lossy, samplerate = sf.read('lossy.wav')
     data_enhanced, samplerate = sf.read('enhanced.wav')