Update app.py

app.py

@@ -6,28 +6,22 @@ import time
 import tempfile
 from huggingface_hub import InferenceClient
 from gtts import gTTS
-
-DESCRIPTION = """ <center><b>JARVIS⚡</b></center> \
-            ### <center>A personal Assistant of Tony Stark for YOU \
-            ### <center>Currently It supports text input, But If this space completes 1k hearts than I starts working on Audio Input.</center> \
-            """
-
-MORE = """ ## TRY Other Models \
-            ### Instant Video: Create Amazing Videos in 5 Second -> https://huggingface.co/spaces/KingNish/Instant-Video \
-            ### Instant Image: 4k images in 5 Second -> https://huggingface.co/spaces/KingNish/Instant-Image \
-            """
-
-Fast = """## Fastest Model"""
-
-Complex = """## Best in Complex Question"""
-
-Detail = """## Best for Detailed Generation or Long Answers"""
-
-client1 = InferenceClient("mistralai/Mixtral-8x7B-Instruct-v0.1")
-
-system_instructions1 = " [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]"
-
-async def generate1(prompt):
+import speech_recognition as sr
+
+# ...
+
+async def generate_audio(prompt):
+    # 음성 인식
+    r = sr.Recognizer()
+    with sr.Microphone() as source:
+        print("Speak:")
+        audio = r.listen(source)
+        try:
+            text = r.recognize_google(audio)
+        except:
+            return "Could not understand audio"
+
+    # LLM 모델에 입력
     generate_kwargs = dict(
         temperature=0.6,
         max_new_tokens=256,
@@ -36,68 +30,21 @@ async def generate1(prompt):
         do_sample=True,
         seed=42,
     )
-    formatted_prompt = system_instructions1 + prompt + "[JARVIS]"
+    formatted_prompt = system_instructions1 + text + "[JARVIS]"
     stream = client1.text_generation(
         formatted_prompt, **generate_kwargs, stream=True, details=True, return_full_text=True)
     output = ""
     for response in stream:
         output += response.token.text
 
+    # 음성 출력
     tts = gTTS(output, lang="ko")
     with tempfile.NamedTemporaryFile(delete=False, suffix=".mp3") as tmp_file:
         tmp_path = tmp_file.name
         tts.save(tmp_path)
-    yield tmp_path
-
-client2 = InferenceClient("meta-llama/Meta-Llama-3-70B-Instruct")
-
-system_instructions2 = "[SYSTEM] Answer as Real Jarvis JARVIS, Made by 'Tony Stark', Must answer in friendly style and Easy Manner. You can answer Complex Questions. Do not say who are you or Hi, Hello, Just Start answering. Stop, as answer ends. [USER]"
-
-async def generate2(prompt):
-    generate_kwargs = dict(
-        temperature=0.6,
-        max_new_tokens=512,
-        top_p=0.95,
-        repetition_penalty=1,
-        do_sample=True,
-    )    
-    formatted_prompt = system_instructions2 + prompt + "[ASSISTANT]"
-    stream = client2.text_generation(
-        formatted_prompt, **generate_kwargs, stream=True, details=True, return_full_text=True)
-    output = ""
-    for response in stream:
-        output += response.token.text
-
-    tts = gTTS(output, lang="ko")
-    with tempfile.NamedTemporaryFile(delete=False, suffix=".mp3") as tmp_file:
-        tmp_path = tmp_file.name
-        tts.save(tmp_path)
-    yield tmp_path
-
-client3 = InferenceClient("meta-llama/Meta-Llama-3-70B-Instruct")
-
-system_instructions3 = "[SYSTEM]The text provided is a request for a specific type of response from me, the virtual assistant. The request asks me to provide detailed and friendly responses as if I am the character Jarvis, inspired 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, my task is to generate responses that appear as if they are created by the user and not an AI-powered assistant.[USER]"
-
-async def generate3(prompt):
-    generate_kwargs = dict(
-        temperature=0.6,
-        max_new_tokens=2048,
-        top_p=0.95,
-        repetition_penalty=1,
-        do_sample=True,
-    )    
-    formatted_prompt = system_instructions3 + prompt + "[ASSISTANT]"
-    stream = client3.text_generation(
-        formatted_prompt, **generate_kwargs, stream=True, details=True, return_full_text=True)
-    output = ""
-    for response in stream:
-        output += response.token.text
+    return tmp_path
 
-    tts = gTTS(output, lang="ko")
-    with tempfile.NamedTemporaryFile(delete=False, suffix=".mp3") as tmp_file:
-        tmp_path = tmp_file.name
-        tts.save(tmp_path)
-    yield tmp_path
+# ...
 
 with gr.Blocks(css="style.css") as demo:    
     gr.Markdown(DESCRIPTION)
@@ -108,12 +55,14 @@ with gr.Blocks(css="style.css") as demo:
                         interactive=False,
                         autoplay=True,
                         elem_classes="audio")
-    with gr.Row():
         translate_btn = gr.Button("Response")
+    with gr.Row():
         translate_btn.click(fn=generate1, inputs=user_input,
-                            outputs=output_audio, api_name="translate")  
-        
-gr.Markdown(MORE)
+                            outputs=output_audio, api_name="translate")
+        translate_btn.click(fn=generate_audio, inputs=user_input,
+                            outputs=output_audio, api_name="generate_audio")
+    with gr.Row():
+        gr.Markdown(MORE)
 
 if __name__ == "__main__":
-    demo.queue(max_size=200).launch()
+    demo.queue(max_size=200).launch()