Update app.py

app.py

@@ -13,18 +13,19 @@ client = InferenceClient("mistralai/Mistral-7B-Instruct-v0.1")
 # Initialize DuckDuckGo search tool
 duckduckgo_search = DuckDuckGoSearchRun()
 
+# Initialize ASR model
 default_lang = "en"
 engines = { default_lang: Model(default_lang) }
 
 def transcribe(audio):
+    """Transcribes the audio file to text."""
     lang = "en"
     model = engines[lang]
     text = model.stt_file(audio)[0]
     return text
 
-system_instructions1 = "<s>[SYSTEM] Answer as Real Jarvis JARVIS, Made by 'Tony Stark', Keep conversation very short, clear, friendly and concise. The text provided is a request for a specific type of response from you, the virtual assistant. The request asks you to provide friendly responses as if You are the character Jarvis, made by 'Tony Stark.' The expectation is that I will avoid introductions and start answering the query directly, elaborating on all aspects of the request. As an AI-powered assistant, your task is to generate responses that appear as if they are created by the user and not an AI-powered assistant. [USER]"
-
 def format_prompt(message, history):
+    """Formats the prompt for the language model."""
     prompt = "<s>"
     for user_prompt, bot_response in history:
         prompt += f"[INST] {user_prompt} [/INST]"
@@ -33,6 +34,7 @@ def format_prompt(message, history):
     return prompt
 
 def generate(prompt, history, temperature=0.9, max_new_tokens=512, top_p=0.95, repetition_penalty=1.0):
+    """Generates a response from the language model."""
     temperature = float(temperature)
     if temperature < 1e-2:
         temperature = 1e-2
@@ -54,10 +56,8 @@ def generate(prompt, history, temperature=0.9, max_new_tokens=512, top_p=0.95, r
 
     for response in stream:
         output += response.token.text
-        # Yield model's response first
         yield output
 
-    # Now, perform DuckDuckGo search and yield results
     search_result = duckduckgo_search.run(prompt)
     if search_result:
         yield search_result
@@ -65,13 +65,26 @@ def generate(prompt, history, temperature=0.9, max_new_tokens=512, top_p=0.95, r
         yield "Sorry, I couldn't find any relevant information."
 
 async def respond(audio):
-    user = transcribe(audio)
-    reply = model(user)
-    communicate = edge_tts.Communicate(reply)
-    with tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as tmp_file:
-        tmp_path = tmp_file.name
-        await communicate.save(tmp_path)
-    yield tmp_path
+    """Handles the full pipeline: transcribe, generate response, and TTS."""
+    try:
+        # Transcribe audio to text
+        user_text = transcribe(audio)
+        
+        # Generate response using the language model
+        history = []
+        response_generator = generate(user_text, history)
+        response_text = ""
+        for response in response_generator:
+            response_text = response
+
+        # Convert the text response to speech
+        communicate = edge_tts.Communicate(response_text)
+        with tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as tmp_file:
+            tmp_path = tmp_file.name
+            await communicate.save(tmp_path)
+        return tmp_path
+    except Exception as e:
+        return str(e)
 
 additional_inputs = [
     gr.Slider(
@@ -124,21 +137,11 @@ with gr.Blocks(css=customCSS) as demo:
     gr.Markdown("Getting real-time updated results for prompts is still proprietary in the face of GPT-4, Co-Pilot etc. This app serves as an open-source alternative for this! UPDATE: Previous version of this app i.e. RAG_FRIDAY_mark_2 has faced some technical issues due to rate limit errors. Problem and solution have been updated by me thanks to this community thread: https://github.com/joaomdmoura/crewAI/issues/136")
 
     with gr.Row():
-        input_audio = gr.Audio(label="Voice Chat (BETA)", sources="microphone", type="filepath", waveform_options=False)
+        input_audio = gr.Audio(label="Voice Chat (BETA)", source="microphone", type="filepath", waveform_options=False)
         output_audio = gr.Audio(label="JARVIS", type="filepath", interactive=False, autoplay=True, elem_classes="audio")
         gr.Interface(fn=respond, inputs=[input_audio], outputs=[output_audio], live=True)
 
     gr.Markdown("## Additional Parameters")
-    with gr.Row():
-        user_input = gr.Textbox(label="Prompt", value="What is Wikipedia")
-        input_text = gr.Textbox(label="Input Text", elem_id="important")
-        output_audio = gr.Audio(label="JARVIS", type="filepath", interactive=False, autoplay=True, elem_classes="audio")
-
-    with gr.Row():
-        translate_btn = gr.Button("Response")
-        translate_btn.click(fn=generate, inputs=user_input, outputs=output_audio, api_name="translate")
-
-    gr.Markdown("## Additional Settings")
     for slider in additional_inputs:
         slider.render()