Update app.py

app.py

@@ -1,10 +1,99 @@
-import gradio as gr 
-import pandas as pd 
+import gradio as gr
+import pandas as pd
 import yt_dlp
 import os
-from transcriber import gen_csv()
-df = pd.read_csv("./final.csv")
-transcripts = df.to_dict(orient='records')
+from semantic_chunkers import StatisticalChunker
+from semantic_router.encoders import HuggingFaceEncoder
+from faster_whisper import WhisperModel
+import spaces
+
+# Function to download YouTube audio
+def download_youtube_audio(url, output_path, preferred_quality="192"):
+    ydl_opts = {
+        'format': 'bestaudio/best',  # Select best audio quality
+        'postprocessors': [{
+            'key': 'FFmpegExtractAudio',
+            'preferredcodec': 'mp3',
+            'preferredquality': preferred_quality,
+        }],
+        'outtmpl': output_path,  # Specify the output path and file name
+    }
+
+    try:
+        with yt_dlp.YoutubeDL(ydl_opts) as ydl:
+            info_dict = ydl.extract_info(url, download=False)
+            video_title = info_dict.get('title', None)
+            print(f"Downloading audio for: {video_title}")
+
+            ydl.download([url])
+            print(f"Audio file saved as: {output_path}")
+
+        return output_path
+
+    except yt_dlp.utils.DownloadError as e:
+        print(f"Error downloading audio: {e}")
+        return None  # Indicate failure
+
+# Function to transcribe audio using WhisperModel
+def transcribe(path, model_name):
+    model = WhisperModel(model_name)
+    print(f"Reading {path}")
+    segments, info = model.transcribe(path)
+    return segments
+
+# Function to process segments and convert them into a DataFrame
+@spaces.GPU
+def process_segments(segments):
+    result = {}
+    print("Processing...")
+    for i, segment in enumerate(segments):
+        chunk_id = f"chunk_{i}"
+        result[chunk_id] = {
+            'chunk_id': segment.id,
+            'chunk_length': segment.end - segment.start,
+            'text': segment.text,
+            'start_time': segment.start,
+            'end_time': segment.end
+        }
+    df = pd.DataFrame.from_dict(result, orient='index')
+    df.to_csv('final.csv')  # Save DataFrame to final.csv
+    return df
+
+# Gradio interface functions
+@spaces.GPU
+def generate_transcript(youtube_url, model_name="distil-large-v3"):
+    path = "downloaded_audio.mp3"
+    download_youtube_audio(youtube_url, path)
+    segments = transcribe(path, model_name)
+    df = process_segments(segments)
+    
+    lis = list(df['text'])
+    encoder = HuggingFaceEncoder(name="sentence-transformers/all-MiniLM-L6-v2")
+    chunker = StatisticalChunker(encoder=encoder, dynamic_threshold=True, min_split_tokens=30, max_split_tokens=40, window_size=2, enable_statistics=False)
+    chunks = chunker._chunk(lis)
+    
+    row_index = 0
+    for i in range(len(chunks)):
+        for j in range(len(chunks[i].splits)):
+            df.at[row_index, 'chunk_id2'] = f'chunk_{i}'
+            row_index += 1
+    
+    grouped = df.groupby('chunk_id2').agg({
+        'start_time': 'min',
+        'end_time': 'max',
+        'text': lambda x: ' '.join(x),
+        'chunk_id': list
+    }).reset_index()
+    
+    grouped = grouped.rename(columns={'chunk_id': 'chunk_ids'})
+    grouped['chunk_length'] = grouped['end_time'] - grouped['start_time']
+    grouped['chunk_id'] = grouped['chunk_id2']
+    grouped = grouped.drop(columns=['chunk_id2', 'chunk_ids'])
+    grouped.to_csv('final.csv')
+    df = pd.read_csv("final.csv")
+    transcripts = df.to_dict(orient='records')
+    
+    return transcripts
 
 # Function to download video using yt-dlp and generate transcript HTML
 def download_video(youtube_url):
@@ -28,6 +117,7 @@ def download_video(youtube_url):
             ydl.download([youtube_url])
 
     # Generate HTML for the transcript
+    transcripts = generate_transcript(youtube_url)
     transcript_html = ""
     for t in transcripts:
         transcript_html += f'<div class="transcript-block"><a href="#" onclick="var video = document.getElementById(\'video-player\').querySelector(\'video\'); video.currentTime={t["start_time"]}; return false;">' \
@@ -37,6 +127,7 @@ def download_video(youtube_url):
 
 # Function to search the transcript
 def search_transcript(keyword):
+    transcripts = pd.read_csv("final.csv").to_dict(orient='records')
     search_results = ""
     for t in transcripts:
         if keyword.lower() in t['text'].lower():
@@ -73,7 +164,6 @@ with gr.Blocks(css=css) as demo:
     # On button click, download the video and display the transcript
     def display_transcript(youtube_url):
         video_path, transcript_html = download_video(youtube_url)
-        # Ensure the video path is correctly passed to the Gradio video component
         return video_path, transcript_html
 
     download_button.click(fn=display_transcript, inputs=youtube_url, outputs=[video, transcript_display])
@@ -82,4 +172,4 @@ with gr.Blocks(css=css) as demo:
     search_button.click(fn=search_transcript, inputs=search_box, outputs=search_results_display)
 
 # Launch the interface
-demo.launch()
+demo.launch()