Update app.py

app.py

@@ -4,6 +4,7 @@ from sklearn.feature_extraction.text import TfidfVectorizer
 from sklearn.metrics.pairwise import cosine_similarity
 import base64
 from pydub import AudioSegment
+from hezar.models import Model
 
 st.set_page_config(
     page_title="Sing It Forward App",
@@ -72,7 +73,6 @@ def cosine_sim(text1, text2):
     return cosine_similarity(vectors)[0, 1]
     
 
-model = whisper.load_model("small")
 
 
 tab1, tab2 = st.tabs(["Take Challenge", "Make Challenge"])
@@ -88,7 +88,8 @@ with tab1:
     if audio_value:
         with open("user_sing.mp3", "wb") as f:
             f.write(audio_value.getbuffer())
-    
+            
+        model = whisper.load_model("base.en")
         user_lyrics = model.transcribe("user_sing.mp3", language="en")["text"]
         st.write(user_lyrics)
         similarity_score = cosine_sim(lyrics, user_lyrics)
@@ -108,7 +109,14 @@ def take_challenge(music_file, typed_lyrics, key, language):
     if audio_value:
         with open("user_sing.mp3", "wb") as f:
             f.write(audio_value.getbuffer())
-        user_lyrics = model.transcribe("user_sing.mp3", language=language)["text"]
+            
+        if language == "en":
+            model = whisper.load_model("base.en")
+            user_lyrics = model.transcribe("user_sing.mp3", language=language)["text"]
+        else:
+            model = Model.load("hezarai/whisper-small-fa")
+            user_lyrics = model.predict("user_sing.mp3")[0]["text"]
+            
         st.write(user_lyrics)
         similarity_score = cosine_sim(typed_lyrics, user_lyrics)
         if similarity_score > 0.85: