Integrate Google Gemini API for conversation generation

app.py

@@ -1,57 +1,102 @@
 import gradio as gr
-from huggingface_hub import InferenceClient
+import google.generativeai as genai
 from gradio_client import Client, handle_file
 import tempfile
 import os
 
+# Configure Gemini API
+GOOGLE_API_KEY = "AIzaSyBN_GLXq2EGdZx9BXMdH58X2ida_y3Ea_M"
+genai.configure(api_key=GOOGLE_API_KEY)
+model = genai.GenerativeModel('gemini-2.0-flash')
+
 # Load TTS client - using the correct Space URL
 tts_client = Client("https://medmac01-darija-arabic-tts.hf.space/")
 
-# Load text generation client
-text_client = InferenceClient("deepseek-ai/DeepSeek-R1-Distill-Qwen-32B")
-
 def generate_conversation(subject, speaker1_audio, speaker2_audio):
-    prompt = f"""
-    Generate a natural Moroccan Darija conversation between two people about: "{subject}".
-    Format:
-    Speaker 1: ...
-    Speaker 2: ...
-    Speaker 1: ...
-    Speaker 2: ...
-    Keep it short and casual (4 lines).
-    """
+    try:
+        prompt = f"""
+        Generate a natural Moroccan Darija conversation in Arabic script only between two people about: "{subject}".
+        Rules:
+        - Use only Arabic script for Darija
+        - Do not include any transliterations or translations
+        - Do not include any Latin characters or parentheses
+        - Use "Speaker 1" and "Speaker 2" as the speaker names
+        
+        Format:
+        Speaker 1: [Arabic Darija text only]
+        Speaker 2: [Arabic Darija text only]
+        Speaker 1: [Arabic Darija text only]
+        Speaker 2: [Arabic Darija text only]
+        
+        Keep it short and casual (4 lines).
+        """
 
-    result = text_client.text_generation(prompt, max_new_tokens=300, temperature=0.7)
-    lines = [line.strip() for line in result.split('\n') if line.strip().startswith("Speaker")]
+        print("Sending prompt to Gemini API...")
+        response = model.generate_content(prompt)
+        print(f"Gemini API Response: {response}")
+        
+        if not response or not response.text:
+            print("No response text received from Gemini API")
+            return ["Error: No response from the model"] + [None] * 4
+            
+        result = response.text
+        print(f"Generated text: {result}")
+        
+        # Split the text into lines and process each line
+        lines = []
+        for line in result.split('\n'):
+            line = line.strip()
+            if ':' in line:
+                # Extract the text after the colon
+                text = line.split(':', 1)[1].strip()
+                # Add the appropriate speaker prefix
+                if len(lines) % 2 == 0:
+                    lines.append(f"Speaker 1: {text}")
+                else:
+                    lines.append(f"Speaker 2: {text}")
+        
+        print(f"Processed lines: {lines}")
+        
+        if not lines:
+            print("No valid lines found in the response")
+            return ["Error: No valid conversation generated"] + [None] * 4
 
-    # Generate audio files using TTS
-    audio_paths = []
-    idx = 0
-    for line in lines:
-        speaker_audio = speaker1_audio if line.startswith("Speaker 1") else speaker2_audio
-        text = line.split(":", 1)[1].strip()
+        # Generate audio files using TTS
+        audio_paths = []
+        idx = 0
+        for line in lines:
+            speaker_audio = speaker1_audio if line.startswith("Speaker 1") else speaker2_audio
+            text = line.split(":", 1)[1].strip()
 
-        # Create TTS audio using the correct API call
-        result = tts_client.predict(
-            text=text,
-            speaker_audio_path=handle_file(speaker_audio),
-            temperature=0.75,
-            api_name="/infer_EGTTS"
-        )
-        
-        # Save the result to a temporary file
-        with tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as tmp:
-            with open(result, "rb") as f:
-                tmp.write(f.read())
-            tmp.flush()
-            audio_paths.append(tmp.name)
-            idx += 1
+            # Create TTS audio using the correct API call
+            result = tts_client.predict(
+                text=text,
+                speaker_audio_path=handle_file(speaker_audio),
+                temperature=0.75,
+                api_name="/infer_EGTTS"
+            )
+            
+            # Save the result to a temporary file
+            with tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as tmp:
+                with open(result, "rb") as f:
+                    tmp.write(f.read())
+                tmp.flush()
+                audio_paths.append(tmp.name)
+                idx += 1
 
-    # Format the conversation text
-    conversation_text = "\n".join(lines)
-    
-    # Return all outputs in the correct order
-    return [conversation_text] + audio_paths
+        # Format the conversation text
+        conversation_text = "\n".join(lines)
+        
+        # Ensure we have exactly 4 audio paths
+        while len(audio_paths) < 4:
+            audio_paths.append(None)
+        
+        # Return all outputs in the correct order
+        return [conversation_text] + audio_paths[:4]
+        
+    except Exception as e:
+        print(f"Error occurred: {str(e)}")
+        return [f"Error: {str(e)}"] + [None] * 4
 
 with gr.Blocks() as demo:
     gr.Markdown("# 🗣️ Moroccan Darija Conversation Generator")