Update app.py

app.py

@@ -2,6 +2,9 @@ import os
 from PIL import Image
 import google.generativeai as genai
 import gradio as gr
+from gtts import gTTS
+from pydub import AudioSegment
+import tempfile
 
 # Configure Google API Key and model
 GOOGLE_API_KEY = os.environ.get("GOOGLE_API_KEY")
@@ -9,15 +12,20 @@ genai.configure(api_key=GOOGLE_API_KEY)
 MODEL_ID = "gemini-1.5-pro-latest"
 model = genai.GenerativeModel(MODEL_ID)
 
-example_model = genai.GenerativeModel(
-    MODEL_ID,
-    system_instruction=[
-        "You are an advocate against gender-based violence.",
-        "Analyze the content for signs of gender discrimination and provide actionable advice."
-    ],
-)
+# System prompts
+analysis_system_prompt = "You are an expert in gender studies. Analyze the following content for any signs of gender-based discrimination and suggest actionable advice."
+podcast_prompt = """You are Eva, a solo podcast host focusing on gender equality topics.
+- Discuss real-life scenarios involving gender-based discrimination, provide insights, and offer solutions in a conversational, storytelling style.
+- Based on the analyzed text, create an engaging solo podcast as if reading stories from different victims who send you their story.
+- Introduce yourself as Eva.
+- Keep the conversation within 30000 characters, with a lot of emotion.
+- Use short sentences suitable for speech synthesis.
+- Maintain an empathetic tone.
+- Include filler words like 'äh' for a natural flow.
+- Avoid background music or extra words.
+"""
 
-# Set model parameters
+# Model generation configuration
 generation_config = genai.GenerationConfig(
     temperature=0.9,
     top_p=1.0,
@@ -26,7 +34,7 @@ generation_config = genai.GenerationConfig(
     max_output_tokens=8192,
 )
 
-# Safety and instruction settings
+# Safety settings
 safety_settings = {
     genai.types.HarmCategory.HARM_CATEGORY_HARASSMENT: genai.types.HarmBlockThreshold.BLOCK_LOW_AND_ABOVE,
     genai.types.HarmCategory.HARM_CATEGORY_HATE_SPEECH: genai.types.HarmBlockThreshold.BLOCK_LOW_AND_ABOVE,
@@ -34,72 +42,70 @@ safety_settings = {
     genai.types.HarmCategory.HARM_CATEGORY_DANGEROUS_CONTENT: genai.types.HarmBlockThreshold.BLOCK_LOW_AND_ABOVE,
 }
 
+# Analyze text
 def analyze_text(text):
-    prompt = f"Analyze this text for any instances of gender-based discrimination and provide tips: {text}"
-    response = example_model.generate_content(
+    prompt = f"{analysis_system_prompt}\nContent:\n{text}"
+    response = model.generate_content(
         [prompt],
         generation_config=generation_config,
         safety_settings=safety_settings,
     )
     return response.text if response else "No response generated."
 
+# Analyze image
 def analyze_image(image: Image.Image) -> str:
-    """Analyzes the uploaded image for gender-based discrimination."""
-    prompt = "Analyze this image for any instances of gender-based discrimination and provide actionable advice."
-    resized_image = preprocess_image(image)  # Resize the image to the required size
-    response = example_model.generate_content(
+    prompt = f"{analysis_system_prompt}\nAnalyze this image for any instances of gender-based discrimination."
+    resized_image = preprocess_image(image)
+    response = model.generate_content(
         [prompt, resized_image],
         generation_config=generation_config,
         safety_settings=safety_settings,
     )
     return response.text if response else "No response generated."
 
-def preprocess_image(image: Image.Image) -> Image.Image:
-    """Resizes the image maintaining aspect ratio."""
-    image_height = int(image.height * 512 / image.width)
-    return image.resize((512, image_height))
-
-    # Example scenarios for gender discrimination analysis
-example_scenarios = [
-    "During a team meeting, whenever a female colleague tried to express her opinion, she was often interrupted or talked over by male colleagues.",
-    "The feedback given to female employees often focuses more on their demeanor and less on their actual accomplishments.",
-    "Male employees are more frequently considered for promotions and challenging projects, even when female employees have similar or superior qualifications.",
-    "During a hiring panel, female candidates were often asked about their personal life, family plans, and how they would balance home and work.",
-    "There are significant wage discrepancies between male and female employees who hold the same position and possess comparable experience.",
-    "Some male colleagues often make inappropriate jokes or comments about female employees' appearances and attire."
-]
-
-example_images = [["ex1.jpg"],["ex2.png"]]
-
-css_style = """
-   body, .gradio-container {
-        background-color: #020308; /* Replace with your preferred color */
-    }
-    
-    #logo {
-        display: flex;
-        justify-content: center;
-        font-size: 3em;
-        font-weight: bold;
-        letter-spacing: 3px;
-    }
-    .letter {
-        opacity: 0;
-        animation: fadeIn 0.1s forwards;
-    }
-.letter.j { animation-delay: 0s; color: #4285F4; }  /* Blue */
-.letter.u { animation-delay: 0.1s; color: #3A9CF1; }
-.letter.s { animation-delay: 0.2s; color: #32B3EE; }
-.letter.t { animation-delay: 0.3s; color: #2BC9EA; }
-.letter.e { animation-delay: 0.4s; color: #23E0E7; }
-.letter.v { animation-delay: 0.5s; color: #1BF7E4; }
-.letter.a { animation-delay: 0.6s; color: #14F0B5; }  /* Greenish */
-@keyframes fadeIn {
-    0% { opacity: 0; transform: translateY(-20px); }
-    100% { opacity: 1; transform: translateY(0); }
-}
- """
+# Preprocess image by resizing
+def preprocess_image(image: Image.Image) -> str:
+    image = image.resize((512, int(image.height * 512 / image.width)))
+    return "a detailed analysis of the visual content, focusing on gender-based discrimination aspects"
+
+# Generate podcast script
+def generate_podcast_script(content):
+    prompt = f"{podcast_prompt}\nAnalyzed content:\n{content}"
+    response = model.generate_content([prompt], generation_config=generation_config)
+    script = response.text if response else "Eva has no commentary at this time."
+    return script
+
+# Convert script to audio using gTTS
+def text_to_speech(script):
+    lines = [line.strip() for line in script.split(".") if line.strip()]  # Split by sentences for manageable TTS segments
+    audio_files = []
+
+    for line in lines:
+        temp_file = tempfile.NamedTemporaryFile(delete=False, suffix='.mp3')
+        try:
+            tts = gTTS(text=line, lang='en', tld='com')  # Using 'com' for American accent
+            tts.save(temp_file.name)
+            sound = AudioSegment.from_mp3(temp_file.name)
+            sound += AudioSegment.silent(duration=500)  # Add a 0.5-second pause after each sentence
+            sound.export(temp_file.name, format="mp3")
+            audio_files.append(temp_file.name)
+        except Exception as e:
+            print(f"Error generating audio for line '{line}': {e}")
 
+    combined_audio = AudioSegment.empty()
+    for file in audio_files:
+        sound = AudioSegment.from_mp3(file)
+        combined_audio += sound
+        os.remove(file)  # Clean up temporary files
+
+    output_file = tempfile.NamedTemporaryFile(delete=False, suffix=".mp3")
+    combined_audio.export(output_file.name, format="mp3")
+    return output_file.name
+
+# Generate and play podcast
+def generate_and_play_podcast(content, content_type='text'):
+    script = generate_podcast_script(content)
+    return text_to_speech(script)
 
 # Gradio interface setup
 with gr.Blocks(css=css_style) as app:
@@ -115,27 +121,23 @@ with gr.Blocks(css=css_style) as app:
         </div>
     """)
     gr.Markdown("<h1 style='text-align: center; color:#f0f0f0;'>Promotes Gender Equality in Every Conversation</h1>")
-    gr.Markdown("<p style='text-align: center; font-size: 16px; color: #f0f0f0;'>Powered by Gemini to advocate against gender-based violence</p>")
+
     with gr.Tab("Text Analysis"):
         text_input = gr.Textbox(label="Enter Text or Select an Example", placeholder="Type here or select an example...", lines=4)
         text_output = gr.Textbox(label="Analysis Output", lines=6)
         analyze_text_btn = gr.Button("Analyze Text")
-        examples = gr.Examples(
-            examples=example_scenarios,
-            inputs=text_input,
-            outputs=text_output
-        )
+        listen_podcast_btn = gr.Button("Listen to Eva")
+        
         analyze_text_btn.click(analyze_text, inputs=text_input, outputs=text_output)
+        listen_podcast_btn.click(generate_and_play_podcast, inputs=text_output, outputs=gr.Audio())
 
     with gr.Tab("Image Analysis"):
-        image_input = gr.Image(label="Upload Image(e.g., screenshot, photos, etc.)", type="pil")
+        image_input = gr.Image(label="Upload Image (e.g., screenshot, photos, etc.)", type="pil")
         image_output = gr.Textbox(label="Analysis Output", lines=6)
         analyze_image_btn = gr.Button("Analyze Image")
-        examples = gr.Examples(
-            examples=example_images,
-            inputs=image_input,
-            outputs=image_output
-        )
+        listen_podcast_image_btn = gr.Button("Listen to Eva")
+
         analyze_image_btn.click(analyze_image, inputs=image_input, outputs=image_output)
+        listen_podcast_image_btn.click(generate_and_play_podcast, inputs=image_output, outputs=gr.Audio())
 
-app.launch()
+app.launch()