Update Milestone5API.py

Milestone5API.py

@@ -3,10 +3,49 @@ import re
 from pytube import YouTube
 from youtube_transcript_api import YouTubeTranscriptApi
 from transformers import MarianMTModel, MarianTokenizer
+import torch
+from TTS.api import TTS
+from moviepy.editor import VideoFileClip
+from pydub import AudioSegment
+def replace_video_audio(video_file_path, new_audio_file_path, output_video_file_path):
+
+    video = VideoFileClip(video_file_path)
+
+    new_audio = AudioSegment.from_wav(new_audio_file_path)
+
+    new_audio = new_audio[:int(video.duration * 1000)]
+
+    video = video.set_audio(new_audio)
+
+    video.write_videofile(output_video_file_path, codec="libx264", audio_codec="aac", temp_audiofile='temp_audio.mp3', remove_temp=True)
+def text_to_speech(translated_text, video_path):
+    device = "cuda" if torch.cuda.is_available() else "gpu"
+    tts = TTS("tts_models/fr/mai/tacotron2-DDC").to(device)
+    text_files_dir = f'CS370_Milestone5/captions/{translated_text}'
+    output_audio_dir = 'CS370_Milestone5/audio'
+    os.makedirs(output_audio_dir, exist_ok=True)
+    
+    for file_name in os.listdir(text_files_dir):
+        if file_name.endswith('.txt'):
+            file_path = os.path.join(text_files_dir, file_name)
+            file_name_only = os.path.splitext(file_name)[0]
+            
+            with open(file_path, 'r', encoding='utf-8') as file:
+                translated_text = file.read()
+                
+            output_file_path = os.path.join(output_audio_dir, f"{file_name_only}.wav")
+            
+            
+            tts.tts_to_file(text=translated_text, file_path=output_file_path)
+
+    output_video_path = f"CS370_Milestone5/videos/{translated_text.split('.')[0]}_new.mp4"
+
+    replace_video_audio(video_path, output_file_path, output_video_path)
+    return output_video_path
 def translate_text_file(input_text_path, output_text_path, source_lang='en', target_lang='fr', model_name="Helsinki-NLP/opus-mt-en-fr", batch_size=8):
     model = MarianMTModel.from_pretrained(model_name)
     tokenizer = MarianTokenizer.from_pretrained(model_name)
-
+    
     def translate_batch(model, tokenizer, sentences):
         sentences = [f"{source_lang}: {sentence}" for sentence in sentences]
         input_ids = tokenizer(sentences, return_tensors="pt", padding=True, truncation=True)["input_ids"]
@@ -75,10 +114,9 @@ def download_video_transcript(video_url, source_lang='en', target_lang='fr'):
                                                       source_lang=source_lang,
                                                       target_lang=target_lang)
 
-            # Delete video file after captions are saved
-            os.remove(video_file)
-            
-            return original_captions, translated_captions, original_filename, f'{captions_path}/{modified_title}_translated.txt'
+            translated_text_filename = f'{captions_path}/{modified_title}_translated.txt'
+            new_video_path = text_to_speech(translated_text_filename, video_file)
+            return original_captions, translated_captions, original_filename, translated_text_filename, new_video_path
         else:
             print("Video ID not found in URL.")
             return None, None, None, None