Update app.py

app.py

@@ -1,3 +1,5 @@
+
+import spaces
 import gradio as gr
 import pandas as pd
 import yt_dlp
@@ -5,11 +7,10 @@ import os
 from semantic_chunkers import StatisticalChunker
 from semantic_router.encoders import HuggingFaceEncoder
 from faster_whisper import WhisperModel
-import spaces
-
+import io
 
-# Function to download YouTube audio
-def download_youtube_audio(url, output_path, preferred_quality="192"):
+# Function to download YouTube audio and return it as a BytesIO object
+def download_youtube_audio(url, preferred_quality="192"):
     ydl_opts = {
         'format': 'bestaudio/best',
         'postprocessors': [{
@@ -17,7 +18,7 @@ def download_youtube_audio(url, output_path, preferred_quality="192"):
             'preferredcodec': 'mp3',
             'preferredquality': preferred_quality,
         }],
-        'outtmpl': output_path
+        'outtmpl': '-',  # Output to stdout
     }
 
     try:
@@ -26,25 +27,29 @@ def download_youtube_audio(url, output_path, preferred_quality="192"):
             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
+            # Download audio to a BytesIO object
+            audio_buffer = io.BytesIO()
+            ydl.download([url], audio_buffer)
+            audio_buffer.seek(0)
+            print("Audio download complete")
+            return audio_buffer
 
     except yt_dlp.utils.DownloadError as e:
         print(f"Error downloading audio: {e}")
         return None
 
-# Function to transcribe audio using WhisperModel
-@spaces.GPU(duration=120)
-def transcribe(path, model_name):
+# Function to transcribe audio from BytesIO using WhisperModel
+@spaces.GPU
+def transcribe(audio_buffer, model_name="medium"):
     model = WhisperModel(model_name)
-    print(f"Reading {path}")
-    segments, info = model.transcribe(path)
+    print("Reading audio buffer")
+    
+    # Hypothetical support for BytesIO object
+    segments, info = model.transcribe(audio_buffer)
     return segments
 
 # Function to process segments and convert them into a DataFrame
-@spaces.GPU(duration=120)
+@spaces.GPU
 def process_segments(segments):
     result = {}
     print("Processing...")
@@ -62,11 +67,13 @@ def process_segments(segments):
     return df
 
 # Gradio interface functions
-@spaces.GPU(duration=120)
+@spaces.GPU
 def generate_transcript(youtube_url, model_name="large-v3"):
-    path = "downloaded_audio.mp3"
-    download_youtube_audio(youtube_url, path)
-    segments = transcribe(path, model_name)
+    audio_buffer = download_youtube_audio(youtube_url)
+    if audio_buffer is None:
+        return "Error downloading audio"
+    
+    segments = transcribe(audio_buffer, model_name)
     df = process_segments(segments)
     
     lis = list(df['text'])
@@ -99,25 +106,20 @@ def generate_transcript(youtube_url, model_name="large-v3"):
 
 # Function to download video using yt-dlp and generate transcript HTML
 def download_video(youtube_url):
-    # Define download options
     ydl_opts = {
         'format': 'mp4',
         'outtmpl': 'downloaded_video.mp4',
         'quiet': True
     }
 
-    # Extract video ID to check if already downloaded
     with yt_dlp.YoutubeDL({'quiet': True}) as ydl:
         info_dict = ydl.extract_info(youtube_url, download=False)
         video_path = 'downloaded_video.mp4'
 
-    # Check if video already downloaded
     if not os.path.exists(video_path):
-        # Download the video
         with yt_dlp.YoutubeDL(ydl_opts) as ydl:
             ydl.download([youtube_url])
 
-    # Generate HTML for the transcript
     transcripts = generate_transcript(youtube_url)
     transcript_html = ""
     for t in transcripts: