Create app.py

app.py

@@ -0,0 +1,91 @@
+import gradio as gr
+import whisper
+from huggingface_hub import InferenceClient
+from gtts import gTTS
+import torch
+import os
+
+# --- 1. YOUR COMPANY DATA (Strict Instructions added) ---
+COMPANY_KNOWLEDGE = """
+You are the reception AI for 'TechNova Solutions'.
+Rules for answering:
+1. Answer ONLY based on the facts below.
+2. Keep answers EXTREMELY SHORT (Maximum 1 sentence).
+3. Be direct. Do not say "Hello" or "Sure" every time. Just give the answer.
+
+FACTS:
+- WHAT WE DO: We build custom solar-powered drones for agriculture.
+- PRICING: Basic drone is $500. Pro drone is $1200.
+- RETURN POLICY: 30-day money-back guarantee.
+- CONTACT: Email help@technova.com or call +1-555-0199.
+- HOURS: Mon-Fri, 9 AM to 5 PM EST.
+"""
+
+# --- 2. SETUP ---
+# Replace with your actual token
+hf_token = os.getenv("HF_TOKEN") or "YOUR_HF_TOKEN_HERE"
+client = InferenceClient(api_key=hf_token)
+
+device = "cuda" if torch.cuda.is_available() else "cpu"
+whisper_model = whisper.load_model("tiny", device=device)
+
+def voice_chat(audio, history):
+    if audio is None:
+        return None, "", history
+
+    # A. Initialize History with Company Knowledge (if empty)
+    if not history:
+        history = [{"role": "system", "content": COMPANY_KNOWLEDGE}]
+
+    try:
+        # B. Hear
+        transcription = whisper_model.transcribe(audio, fp16=False)["text"]
+        
+        # C. Think
+        # We append the user text
+        history.append({"role": "user", "content": transcription})
+        
+        response = client.chat.completions.create(
+            model="HuggingFaceH4/zephyr-7b-beta", 
+            messages=history, 
+            max_tokens=60, # Reduced to 60 to force brevity
+            temperature=0.5 # Lower temperature = more precise/less creative
+        )
+        ai_text = response.choices[0].message.content
+
+        history.append({"role": "assistant", "content": ai_text})
+
+        # D. Speak
+        tts = gTTS(text=ai_text, lang='en')
+        audio_path = "response.mp3"
+        tts.save(audio_path)
+        
+        return audio_path, ai_text, history
+
+    except Exception as e:
+        return None, f"Error: {str(e)}", history
+
+# --- 3. INTERFACE ---
+with gr.Blocks(theme="soft") as demo:
+    gr.Markdown("## 🏢 Quick-Response Company AI")
+    
+    conversation_history = gr.State([])
+
+    with gr.Row():
+        input_audio = gr.Audio(sources=["microphone"], type="filepath", label="Ask Question")
+    
+    with gr.Row():
+        output_audio = gr.Audio(label="AI Response", autoplay=True)
+        output_text = gr.Textbox(label="Transcript")
+    
+    clear_btn = gr.Button("Reset")
+    
+    input_audio.change(
+        voice_chat, 
+        inputs=[input_audio, conversation_history], 
+        outputs=[output_audio, output_text, conversation_history]
+    )
+    
+    clear_btn.click(lambda: ([], None, ""), outputs=[conversation_history, output_audio, output_text])
+
+demo.launch(debug=True)