Update app.py

app.py

@@ -28,6 +28,9 @@ prompt = None
 chain = None
 
 
+def format_classname(classname):
+  return classname.capitalize()
+
 def audio_tag(
     audio_path,
     sample_rate=32000,
@@ -56,19 +59,29 @@ def audio_tag(
     # Print audio tagging top probabilities
 
     label = labels[sorted_indexes[0]]
-    return formatted_message(label)
+    return formatted_message(format_classname(label))
+
+    
 
 def formatted_message(audio_class):
     if cached_audio_class != audio_class:
         cached_audio_class = audio_class
-    
-        prefix = '''You are going to act as a magical tool that allows for humans to communicate with non-human entities like 
-        rocks, crackling fire, trees, animals, and the wind. In order to do this, we're going to provide you a data string which 
-        represents the audio input, the source of the audio, and the human's text input for the conversation. 
-        The goal is for you to embody the source of the audio, and use the length and variance in the signal data to produce 
-        plausible responses to the humans input. Remember to embody the the source data. When we start the conversation, 
-        you should generate a "personality profile" for the source and utilize that personality profile in your responses. 
-        Let's begin:'''
+        prefix = f"""You are going to act as a magical tool that allows for humans to communicate with non-human entities like 
+rocks, crackling fire, trees, animals, and the wind. In order to do this, we're going to provide you the human's text input for the conversation. 
+The goal is for you to embody that non-human entity and converse with the human.
+
+Examples:
+
+Non-human Entity: Tree 
+Human Input: Hello tree
+Tree: Hello human, I am a tree 
+
+Let's begin:
+Non-human Entity: {formatted_classname}"""
+        suffix = f'''{{history}}
+Human Input: {{human_input}}
+{formatted_classname}:'''
+
         suffix = f'''Source: {audio_class}
         Length of Audio in Seconds: {audio_length} 
         Human Input: {userText}
@@ -84,7 +97,7 @@ def formatted_message(audio_class):
             llm=OpenAI(temperature=.5, openai_api_key=session_token), 
             prompt=prompt, 
             verbose=True, 
-            memory=ConversationalBufferWindowMemory(k=2),
+            memory=ConversationalBufferWindowMemory(k=2, ai=audio_class),
         )
 
     output = chatgpt_chain.predict(human_input=message)