Create app.py

app.py

@@ -0,0 +1,93 @@
+import os
+import yt_dlp
+import gradio as gr
+from faster_whisper import WhisperModel
+from transformers import pipeline, AutoModelForSeq2SeqLM, AutoTokenizer
+import openai
+import torch
+
+# Optional: Set your OpenAI API key (use env var for security)
+openai.api_key = "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"  # Replace with your real key or set as os.environ["OPENAI_API_KEY"]
+
+def download_and_extract_audio(youtube_url):
+    output_path = "downloads"
+    os.makedirs(output_path, exist_ok=True)
+
+    ydl_opts = {
+        'format': 'bestaudio/best',
+        'outtmpl': os.path.join(output_path, '%(id)s.%(ext)s'),
+        'postprocessors': [{
+            'key': 'FFmpegExtractAudio',
+            'preferredcodec': 'mp3',
+            'preferredquality': '192',
+        }],
+    }
+
+    with yt_dlp.YoutubeDL(ydl_opts) as ydl:
+        info_dict = ydl.extract_info(youtube_url, download=True)
+        video_id = info_dict.get("id", None)
+        filename = os.path.join(output_path, f"{video_id}.mp3")
+        return filename
+
+def transcribe_audio(audio_path):
+    model = WhisperModel("base", compute_type="int8", device="cuda" if torch.cuda.is_available() else "cpu")
+    segments, _ = model.transcribe(audio_path)
+    transcript = " ".join([seg.text for seg in segments])
+    return transcript
+
+# Preload FLAN-T5 model offline
+tokenizer = AutoTokenizer.from_pretrained("google/flan-t5-large")
+model = AutoModelForSeq2SeqLM.from_pretrained("google/flan-t5-large")
+local_gen = pipeline("text2text-generation", model=model, tokenizer=tokenizer)
+
+def generate_response(transcript, user_prompt, use_online=False):
+    prompt = f"""You are a helpful AI assistant. Based on the transcript of a video, please {user_prompt.strip().lower()}.
+
+Transcript:
+{transcript[:3000]}"""
+
+    if use_online:
+        try:
+            response = openai.ChatCompletion.create(
+                model="gpt-4",
+                messages=[{"role": "user", "content": prompt}],
+                max_tokens=1000
+            )
+            return response.choices[0].message["content"]
+        except Exception as e:
+            return f"⚠️ Online API failed: {str(e)}"
+    else:
+        result = local_gen(prompt, max_length=1024, do_sample=False)
+        return result[0]['generated_text']
+
+def enhanced_ai_study_pipeline(video_source, youtube_url, upload_file, user_prompt, use_online_api):
+    try:
+        if video_source == "YouTube URL":
+            audio_path = download_and_extract_audio(youtube_url)
+        elif video_source == "Upload File" and upload_file is not None:
+            audio_path = upload_file.name
+        else:
+            return "No valid input provided.", ""
+
+        transcript = transcribe_audio(audio_path)
+        ai_response = generate_response(transcript, user_prompt, use_online=use_online_api)
+        return transcript, ai_response
+    except Exception as e:
+        return "Error occurred", str(e)
+
+video_input = gr.Radio(["YouTube URL", "Upload File"], label="Video Source")
+youtube_url = gr.Textbox(label="Enter YouTube URL")
+upload_file = gr.File(label="Upload a Video File", file_types=[".mp4", ".mp3", ".wav"])
+user_prompt = gr.Textbox(label="What do you want from the transcript?", placeholder="e.g., Prepare a diet plan based on this video")
+use_online_api = gr.Checkbox(label="Use Online API (GPT-4)", value=False)
+
+gr.Interface(
+    fn=enhanced_ai_study_pipeline,
+    inputs=[video_input, youtube_url, upload_file, user_prompt, use_online_api],
+    outputs=[
+        gr.Textbox(label="Transcript"),
+        gr.Textbox(label="AI Response")
+    ],
+    title="📚 AI Transcription Assistant (Offline + Online GPT)",
+    description="Upload or paste a YouTube video. Enter your goal and get a smart AI answer. Works offline with FLAN-T5 or online with GPT-4."
+).launch(share=True)