Update app.py

app.py

@@ -10,11 +10,14 @@ import os
 from bs4 import BeautifulSoup
 import re
 import numpy as np
-from moviepy import VideoFileClip
+from moviepy.editor import VideoFileClip
 import soundfile as sf
+from spellchecker import SpellChecker
+import tempfile
 
 # Load the transcription model
 transcription_pipeline = pipeline("automatic-speech-recognition", model="facebook/wav2vec2-base-960h")
+spell = SpellChecker()
 
 def download_audio_from_url(url):
     try:
@@ -40,34 +43,50 @@ def download_audio_from_url(url):
         print(f"Error in download_audio_from_url: {str(e)}")
         raise
 
+def correct_spelling(text):
+    words = text.split()
+    corrected_words = [spell.correction(word) or word for word in words]
+    return ' '.join(corrected_words)
+
+def format_transcript(transcript):
+    sentences = transcript.split('.')
+    formatted_transcript = []
+    current_speaker = None
+    for sentence in sentences:
+        if ':' in sentence:
+            speaker, content = sentence.split(':', 1)
+            if speaker != current_speaker:
+                formatted_transcript.append(f"\n\n{speaker.strip()}:{content.strip()}.")
+                current_speaker = speaker
+            else:
+                formatted_transcript.append(f"{content.strip()}.")
+        else:
+            formatted_transcript.append(sentence.strip() + '.')
+    return ' '.join(formatted_transcript)
+
 def transcribe_audio(video_bytes):
     try:
-        # Save the video bytes to a temporary file
         with open("temp_video.mp4", "wb") as f:
             f.write(video_bytes)
 
-        # Extract audio from video
         video = VideoFileClip("temp_video.mp4")
         audio = video.audio
 
-        # Export audio as mono WAV
         audio.write_audiofile("temp_audio.wav", fps=16000, nbytes=2, codec='pcm_s16le')
 
-        # Load the audio file
         audio_data, sample_rate = sf.read("temp_audio.wav")
 
-        # Ensure audio is mono
         if len(audio_data.shape) > 1:
             audio_data = audio_data.mean(axis=1)
 
-        # Normalize the audio data
         audio_data = audio_data.astype(np.float32) / np.max(np.abs(audio_data))
 
-        # Transcribe the audio
         result = transcription_pipeline(audio_data)
         transcript = result['text']
 
-        # Clean up temporary files
+        transcript = correct_spelling(transcript)
+        transcript = format_transcript(transcript)
+
         os.remove("temp_video.mp4")
         os.remove("temp_audio.wav")
 
@@ -91,9 +110,12 @@ def transcribe_video(url):
         error_message = f"An error occurred: {str(e)}"
         print(error_message)
         return error_message
-        
+
 def download_transcript(transcript):
-    return transcript, "transcript.txt"
+    with tempfile.NamedTemporaryFile(mode='w', delete=False, suffix='.txt') as temp_file:
+        temp_file.write(transcript)
+        temp_file_path = temp_file.name
+    return temp_file_path
 
 # Create the Gradio interface
 with gr.Blocks(title="Video Transcription") as demo:
@@ -105,6 +127,6 @@ with gr.Blocks(title="Video Transcription") as demo:
     download_link = gr.File(label="Download Transcript")
     
     transcribe_button.click(fn=transcribe_video, inputs=video_url, outputs=transcript_output)
-    download_button.click(fn=download_transcript, inputs=transcript_output, outputs=[download_link, download_link])
+    download_button.click(fn=download_transcript, inputs=transcript_output, outputs=download_link)
 
 demo.launch()