[FEATURE] enhance YouTube transcript summarization with dynamic chunking and API key input

app.py

@@ -2,9 +2,10 @@ import gradio as gr
 from tool import YouTubeTranscriptExtractor, TranscriptSummarizer
 
 youtube_tool = YouTubeTranscriptExtractor()
-summarizer_tool = TranscriptSummarizer()
+#summarizer_tool = TranscriptSummarizer()
 
-def process_youtube_video(video_url):
+def process_youtube_video(video_url, hf_api_key):
+    summarizer_tool = TranscriptSummarizer(hf_api_key=hf_api_key)
     transcript = youtube_tool.forward(video_url=video_url)
     summary_and_blog = summarizer_tool.forward(transcript=transcript)
     try:
@@ -16,14 +17,17 @@ def process_youtube_video(video_url):
 
 iface = gr.Interface(
     fn=process_youtube_video,
-    inputs="text",
+    inputs=[
+        gr.Textbox(label="YouTube Video URL"),
+        gr.Textbox(label="Hugging Face API Key", type="password")
+    ],
     outputs=[
         gr.Textbox(label="Transcript"),
         gr.Textbox(label="Summary and Blog Content"),
         gr.Image(label="Generated Image", image_mode="RGBA")
     ],
     title="YouTube Transcript Summarizer and Blog Content Generator",
-    description="Enter a YouTube video URL to extract the transcript, summarize it, and generate blog content with an image."
+    description="Enter a YouTube video URL and Hugging Face API Key to extract the transcript, summarize it, and generate blog content with an image."
 )
 
 iface.launch()

requirements.txt

@@ -4,6 +4,5 @@ gradio
 transformers
 torch
 torchvision
-python-dotenv
 requests
 Pillow

tool.py

@@ -5,9 +5,9 @@ from transformers import pipeline
 import requests
 import io
 from PIL import Image
-from dotenv import load_dotenv
+#from dotenv import load_dotenv
 
-load_dotenv()
+#load_dotenv()
 
 class TranscriptSummarizer(Tool):
     description = "Summarizes a transcript and generates blog content using the transformers library and Hugging Face API for image generation."
@@ -15,11 +15,12 @@ class TranscriptSummarizer(Tool):
     inputs = {'transcript': {'type': 'string', 'description': 'The transcript to summarize.'}}
     output_type = "string"
 
-    def __init__(self, *args, **kwargs):
+    def __init__(self, *args, hf_api_key: str = None, **kwargs):
         super().__init__(*args, **kwargs)
         self.summarizer = pipeline("summarization", model="google/pegasus-xsum")
         self.api_url = "https://api-inference.huggingface.co/models/ZB-Tech/Text-to-Image"
-        self.headers = {"Authorization": f"Bearer {os.getenv('HF_API_KEY')}"}
+        self.hf_api_key = hf_api_key
+        self.headers = {"Authorization": f"Bearer {self.hf_api_key}"}
 
     def query(self, payload):
         response = requests.post(self.api_url, headers=self.headers, json=payload)
@@ -27,23 +28,34 @@ class TranscriptSummarizer(Tool):
 
     def forward(self, transcript: str) -> str:
         try:
+            if not self.hf_api_key:
+                return "Hugging Face API key is required. Please provide it in the input field."
+
             transcript_length = len(transcript)
 
             def get_summary_lengths(length):
-                if length <= 1000:
-                    max_length = 300
-                    min_length = 100
-                elif length <= 3000:
-                    max_length = 750
-                    min_length = 250
-                else:
-                    max_length = 1500
-                    min_length = 500
+                # set the short maths formula
+                max_length = int(length * 0.8)
+                min_length = int(length * 0.2)
                 return max_length, min_length
 
-            max_length, min_length = get_summary_lengths(transcript_length)
-            summary = self.summarizer(transcript, max_length=max_length, min_length=min_length, do_sample=False)[0]['summary_text']
-            key_entities = summary.split()[:3]  # Extract first 3 words as key entities
+            # Split the transcript into chunks of 500 characters make it dynamic according to the length of the transcript
+            if transcript_length < 500:
+                return "Transcript is too short to summarize."
+            chunk_size = 500
+            transcript_chunks = [transcript[i:i+chunk_size] for i in range(0, len(transcript), chunk_size)]
+
+            # Summarize each chunk of the transcript
+            summaries = []
+            for chunk in transcript_chunks:
+                max_length, min_length = get_summary_lengths(len(chunk))
+                summary = self.summarizer(chunk, max_length=max_length, min_length=min_length, do_sample=False)[0]['summary_text']
+                summaries.append(summary)
+
+            # Concatenate the summaries
+            full_summary = "\n".join(summaries)
+
+            key_entities = full_summary.split()[:15]  # Extract first 3 words as key entities
             image_prompt = f"Generate an image related to: {' '.join(key_entities)}, cartoon style"
             image_bytes = self.query({"inputs": image_prompt})
             image = Image.open(io.BytesIO(image_bytes))
@@ -52,7 +64,7 @@ class TranscriptSummarizer(Tool):
                 os.makedirs(image_folder)
             image_url = os.path.join(image_folder, "image.jpg")  # Specify the folder path
             image.save(image_url) # Save the image to a file
-            return f"{summary}\n\nImage URL: {image_url}" # Return the file path
+            return f"{full_summary}\n\nImage URL: {image_url}" # Return the file path
         except Exception as e:
             return f"An unexpected error occurred: {str(e)}"