Update app.py

app.py

@@ -24,6 +24,9 @@ from PLCMOS.plc_mos import PLCMOSEstimator
 from speechmos import dnsmos
 from speechmos import plcmos
 
+import speech_recognition as sr
+from jiwer import wer
+
 
 @st.cache
 def load_model():
@@ -272,10 +275,10 @@ if st.button('Сгенерировать потери'):
 
     PLC_massv2 = [plcmos.run("target.wav", sr=16000)['plcmos'], plcmos.run("lossy.wav", sr=16000)['plcmos'], plcmos.run("enhanced.wav", sr=16000)['plcmos']]
 
-    DNS = [dnsmos.run("target.wav", sr=16000)['ovrl_mos'], dnsmos.run("lossy.wav", sr=16000)['ovrl_mos'], dnsmos.run("enhanced.wav", sr=16000)['ovrl_mos']]
+    #DNS = [dnsmos.run("target.wav", sr=16000)['ovrl_mos'], dnsmos.run("lossy.wav", sr=16000)['ovrl_mos'], dnsmos.run("enhanced.wav", sr=16000)['ovrl_mos']]
 
     df_1['PLCMOSv2'] = PLC_massv2
-    df_1['DNSMOS'] = DNS
+    #df_1['DNSMOS'] = DNS
     
 
     #df_2 = pd.DataFrame(columns=['DNSMOS', 'PLCMOSv2'])
@@ -290,6 +293,33 @@ if st.button('Сгенерировать потери'):
     #df_2.columns = new_columns
     #df_merged = df_1.merge(df_2, left_index=True, right_index=True)
 
+
+    r = sr.Recognizer ()
+
+    harvard = sr.AudioFile('target.wav')
+    with harvard as source:
+        audio = r.record(source)
+
+    orig = r.recognize_google(audio, language = "ru-RU")
+
+    harvard = sr.AudioFile('lossy.wav')
+    with harvard as source:
+        audio = r.record(source)
+    lossy = r.recognize_google(audio, language = "ru-RU")
+
+
+    harvard = sr.AudioFile('enhanced.wav')
+    with harvard as source:
+        audio = r.record(source)
+    enhanced = r.recognize_google(audio, language = "ru-RU")
+
+    error1 = wer(orig, orig)
+    error2 = wer(orig, lossy)
+    error2 = wer(orig, enhanced)
+    WER_mass=[error1, error2, error3]
+
+    df_1['WER'] = WER_mass
+    
     st.dataframe(df_1)