Update app.py

app.py

@@ -6,17 +6,24 @@ from gtts import gTTS
 from transformers import pipeline
 from huggingface_hub import InferenceClient
 
+# Model names
 ASR_MODEL_NAME = "openai/whisper-small"
 LLM_MODEL_NAME = "mistralai/Mistral-7B-Instruct-v0.2"
 
+# Initial system prompt
 system_prompt = """"<s>[INST] You are Friday, a helpful and conversational AI assistant, and you respond with one to two sentences. [/INST] Hello there! I'm Friday, how can I help you?</s>"""
 
-instruct_history = system_prompt + """"""
+# Global variables for history
+instruct_history = system_prompt
+formatted_history = ""
 
+# Create inference client for text generation
 client = InferenceClient(LLM_MODEL_NAME)
 
+# Set device for ASR pipeline
 device = 0 if torch.cuda.is_available() else "cpu"
 
+# ASR pipeline
 pipe = pipeline(
     task="automatic-speech-recognition",
     model=ASR_MODEL_NAME,
@@ -44,7 +51,9 @@ def generate(instruct_history, temperature=0.1, max_new_tokens=128, top_p=0.95,
     return output
 
 @spaces.GPU(duration=60)
-def transcribe(audio, instruct_history=instruct_history):
+def transcribe(audio):
+    global instruct_history, formatted_history
+
     sr, y = audio
     y = y.astype(np.float32)
     y /= np.max(np.abs(y))
@@ -53,27 +62,25 @@ def transcribe(audio, instruct_history=instruct_history):
     transcribed_user_audio = pipe({"sampling_rate": sr, "raw": y})["text"]
 
     # Append user input to history
-    formatted_history = f"""Human: {transcribed_user_audio}\n\n"""
-    instruct_history += f"""<s>[INST] {transcribed_user_audio} [/INST] """
+    formatted_history += f"😃 Human: {transcribed_user_audio}\n\n"
+    instruct_history += f"<s>[INST] {transcribed_user_audio} [/INST] "
 
     # Generate LLM response
     llm_response = generate(instruct_history)
 
     # Append AI response to history
-    instruct_history += f""" {llm_response}</s>"""
-    formatted_history += f"""Friday: {llm_response}\n\n"""
+    instruct_history += f" {llm_response}</s>"
+    formatted_history += f"🤖 Friday: {llm_response}\n\n"
 
     # Convert AI response to audio
     audio_response = gTTS(llm_response)
     audio_response.save("response.mp3")
 
-    # Display the full conversation history
-    full_history = formatted_history
-
-    return "response.mp3", full_history
+    # Return the full conversation history
+    return "response.mp3", formatted_history
 
 with gr.Blocks() as demo:
-    gr.HTML("<center><h1>Friday: AI Virtual Assistant</h1><center>")
+    gr.HTML("<center><h1>Friday: AI Virtual Assistant 🤖</h1><center>")
 
     with gr.Row():
         audio_input = gr.Audio(label="Human", sources="microphone")