Update functions.py

functions.py

@@ -70,7 +70,7 @@ output_parser = RegexParser(
 )
 
 system_template="""Use only the following pieces of finance context to answer the users question thoroughly.
-Do not use any information not provided in the context. 
+Do not use any information not provided in the context and remember you are a finance expert.
 If you don't know the answer, just say that you don't know, don't try to make up an answer.
 ALWAYS return a "SOURCES" part in your answer.
 The "SOURCES" part should be a reference to the source of the document from which you got your answer.
@@ -126,6 +126,13 @@ def load_asr_model(asr_model_name):
     
     return asr_model
 
+@st.experimental_singleton(suppress_st_warning=True)
+def load_whisper_api(audio):
+    file = open(audio, "rb")
+    transcript = openai.Audio.translate("whisper-1", file)
+
+    return transcript
+
 @st.experimental_singleton(suppress_st_warning=True)
 def process_corpus(corpus, title, embedding_model, chunk_size=1000, overlap=50):
 
@@ -217,54 +224,87 @@ def get_spacy():
 @st.experimental_memo(suppress_st_warning=True)
 def inference(link, upload, _asr_model):
     '''Convert Youtube video or Audio upload to text'''
+
+    try:
+        
+        if validators.url(link):
+        
+            yt = YouTube(link)
+            title = yt.title
     
-    if validators.url(link):
+            #Get audio file from YT
+            audio_file = yt.streams.filter(only_audio=True)[0].download(filename="audio.mp4")
     
-        yt = YouTube(link)
-        title = yt.title
-
-        #Get audio file from YT
-        audio_file = yt.streams.filter(only_audio=True)[0].download(filename="audio.mp4")
-
-        #Get size of audio file
-        audio_size = round(os.path.getsize(path)/(1024*1024),1)
-
-        #Check if file is > 24mb, if not then use Whisper API
-        if audio_size <= 24:
-
-            #Use whisper API
-            transcript = openai.Audio.translate("whisper-1", audio_file)
-
-        else:
+            #Get size of audio file
+            audio_size = round(os.path.getsize(path)/(1024*1024),1)
+    
+            #Check if file is > 24mb, if not then use Whisper API
+            if audio_size <= 25:
+    
+                #Use whisper API
+                results = load_whisper_api(audio_file)['text']
+    
+            else:
+    
+                st.write('File size larger than 24mb, applying chunking and transcription')
+    
+                song = AudioSegment.from_file("audio.mp4", format='mp4')
+    
+                # PyDub handles time in milliseconds
+                twenty_minutes = 20 * 60 * 1000
+                
+                chunks = song[::twenty_minutes]
+                
+                transcriptions = []
+                
+                for i, chunk in enumerate(chunks):
+                    chunk.export(f'output/chunk_{i}.mp4', format='mp4')
+                    transcriptions.append(load_whisper_api('output/chunk_{i}.mp4')['text'])
+    
+                results = ','.join(transcriptions)
 
-            st.write('File size larger than 24mb, applying chunking and transcription')
+            return results, yt.title
 
-            # load the audio file
-            audio_file = AudioSegment.from_file(path, format="mp4")
-            
-            # set chunk size to 24mb (in bytes)
-            chunk_size = 24 * 1024 * 1024
+        elif upload:
 
-            # create a directory to store the output files
-            if not os.path.exists("audio_output"):
-                os.mkdir("audio_output")
+            #Get size of audio file
+            audio_size = round(os.path.getsize(path)/(1024*1024),1)
+    
+            #Check if file is > 24mb, if not then use Whisper API
+            if audio_size <= 25:
+    
+                #Use whisper API
+                results = load_whisper_api(audio_file)['text']
+    
+            else:
+    
+                st.write('File size larger than 24mb, applying chunking and transcription')
+    
+                song = AudioSegment.from_file("audio.mp4", format='mp4')
+    
+                # PyDub handles time in milliseconds
+                twenty_minutes = 20 * 60 * 1000
+                
+                chunks = song[::twenty_minutes]
+                
+                transcriptions = []
+                
+                for i, chunk in enumerate(chunks):
+                    chunk.export(f'output/chunk_{i}.mp4', format='mp4')
+                    transcriptions.append(load_whisper_api('output/chunk_{i}.mp4')['text'])
+    
+                results = ','.join(transcriptions)
 
-            audio_chunks = []
-            
-            # iterate over each chunk and export it as a separate file
-            for i, chunk in enumerate(audio_file[::chunk_size]):
-                chunk.export(f"output/chunk_{i}.mp4", format="mp4")
-                audio_chunks.append(f"output/chunk_{i}.mp4")
+            return results, "Transcribed Earnings Audio"
                 
+    except:
 
-      results = _asr_model.transcribe(path, task='transcribe', language='en')
-      
-      return results['text'], yt.title
-      
-    elif upload:
-      results = _asr_model.trasncribe(upload, task='transcribe', language='en')
+        st.write('Whisper API Error, using Whisper module from GitHub, might take longer than expected')
+        
+        results = _asr_model.transcribe(path, task='transcribe', language='en')
       
-      return results['text'], "Transcribed Earnings Audio"
+        return results['text'], yt.title
+    
       
 @st.experimental_memo(suppress_st_warning=True)
 def sentiment_pipe(earnings_text):