Create app.py

app.py

@@ -0,0 +1,71 @@
+import spaces
+import tempfile
+import asyncio
+import gradio as gr
+from streaming_stt_nemo import Model
+from huggingface_hub import InferenceClient
+import edge_tts
+
+default_lang = "en"
+engines = {default_lang: Model(default_lang)}
+
+def transcribe(audio):
+    lang = "en"
+    model = engines[lang]
+    text = model.stt_file(audio)[0]
+    return text
+
+client = InferenceClient("mistralai/Mixtral-8x7B-Instruct-v0.1")
+
+system_instructions = "[SYSTEM] Answer as Real OpenGPT 4o, Made by 'KingNish', 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. You 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 model(text):
+    generate_kwargs = dict(
+        temperature=0.7,
+        max_new_tokens=512,
+        top_p=0.95,
+        repetition_penalty=1,
+        do_sample=True,
+        seed=42,
+    )
+    
+    formatted_prompt = system_instructions + text + "[OpenGPT 4o]"
+    stream = client.text_generation(
+        formatted_prompt, **generate_kwargs, stream=True, details=True, return_full_text=False)
+    output = ""
+    for response in stream:
+        if not response.token.text == "</s>":
+            output += response.token.text
+
+    return output
+
+@spaces.GPU(duration=120)  # Increase duration if needed
+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)
+    return tmp_path
+
+with gr.Blocks() as voice:   
+    with gr.Row():
+        input = gr.Audio(label="Voice Chat", source="microphone", type="filepath")
+        output = gr.Audio(label="OpenGPT 4o", type="filepath", interactive=False, autoplay=True)
+    
+    gr.Interface(
+        fn=respond,
+        inputs=[input],
+        outputs=[output],
+        live=True,
+    )
+
+theme = gr.themes.Base()
+
+with gr.Blocks(theme=theme, css="footer {visibility: hidden}textbox{resize:none}", title="GPT 4o DEMO") as demo:
+    gr.Markdown("# OpenGPT 4o")
+    gr.TabbedInterface([voice], ['🗣️ Voice Chat'])
+
+demo.queue(max_size=200)
+demo.launch()