Update test3.py

test3.py

@@ -12,13 +12,14 @@ import shutil
 import srt
 from tqdm import tqdm
 import concurrent.futures
+import gc
 
 mtypes = {"cpu": "int8", "cuda": "float16"}
 
 def setup_logging():
     logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(levelname)s - %(message)s')
 
-def process_audio_file(audio_file, args):
+def process_audio_file(audio_file, args, whisper_model, alignment_model, metadata):
     logging.info(f"Processing file: {audio_file}")
     
     if args.stemming:
@@ -48,8 +49,7 @@ def process_audio_file(audio_file, args):
             vocal_target,
             args.language,
             args.batch_size,
-            args.model_name,
-            mtypes[args.device],
+            whisper_model,
             args.suppress_numerals,
             args.device,
         )
@@ -58,17 +58,13 @@ def process_audio_file(audio_file, args):
         whisper_results, language = transcribe(
             vocal_target,
             args.language,
-            args.model_name,
-            mtypes[args.device],
+            whisper_model,
             args.suppress_numerals,
             args.device,
         )
 
     logging.info("Aligning transcription...")
     if language in wav2vec2_langs:
-        alignment_model, metadata = whisperx.load_align_model(
-            language_code=language, device=args.device
-        )
         result_aligned = whisperx.align(
             whisper_results, alignment_model, metadata, vocal_target, args.device
         )
@@ -77,8 +73,6 @@ def process_audio_file(audio_file, args):
             initial_timestamp=whisper_results[0].get("start"),
             final_timestamp=whisper_results[-1].get("end"),
         )
-        del alignment_model
-        torch.cuda.empty_cache()
     else:
         word_timestamps = []
         for segment in whisper_results:
@@ -91,7 +85,8 @@ def process_audio_file(audio_file, args):
     ROOT = os.getcwd()
     temp_path = os.path.join(ROOT, "temp_outputs")
     os.makedirs(temp_path, exist_ok=True)
-    sound.export(os.path.join(temp_path, "mono_file.wav"), format="wav")
+    mono_file = os.path.join(temp_path, "mono_file.wav")
+    sound.export(mono_file, format="wav")
 
     # Initialize NeMo MSDD diarization model
     logging.info("Performing diarization...")
@@ -125,9 +120,6 @@ def process_audio_file(audio_file, args):
     audio_dir = os.path.join(autodiarization_dir, base_name)
     os.makedirs(audio_dir, exist_ok=True)
 
-    # Create a dictionary to store speaker-specific metadata
-    speaker_metadata = {}
-
     # Generate the SRT file
     srt_file = f"{os.path.splitext(audio_file)[0]}.srt"
     with open(srt_file, "w", encoding="utf-8") as f:
@@ -142,7 +134,7 @@ def process_audio_file(audio_file, args):
 
     # Process each segment in the SRT data
     logging.info("Processing audio segments...")
-    for segment in tqdm(srt_segments, desc="Processing segments"):
+    for segment in tqdm(srt_segments, desc="Processing segments", leave=False):
         start_time = segment.start.total_seconds() * 1000
         end_time = segment.end.total_seconds() * 1000
         speaker_name, transcript = segment.content.split(": ", 1)
@@ -156,21 +148,33 @@ def process_audio_file(audio_file, args):
         os.makedirs(os.path.dirname(segment_path), exist_ok=True)
         segment_audio.export(segment_path, format="wav")
 
-        # Store the metadata for each speaker
-        if speaker_name not in speaker_metadata:
-            speaker_metadata[speaker_name] = []
-        speaker_metadata[speaker_name].append(f"speaker_{speaker_id}_{segment.index:03d}|{speaker_name}|{transcript}")
-
-    # Write the metadata.csv file for each speaker
-    for speaker_name, metadata in speaker_metadata.items():
-        speaker_id = int(speaker_name.split(" ")[-1])
+        # Write metadata directly to file
         speaker_dir = os.path.join(audio_dir, f"speaker_{speaker_id}")
-        with open(os.path.join(speaker_dir, "metadata.csv"), "w", encoding="utf-8") as f:
-            f.write("\n".join(metadata))
+        with open(os.path.join(speaker_dir, "metadata.csv"), "a", encoding="utf-8") as f:
+            f.write(f"speaker_{speaker_id}_{segment.index:03d}|{speaker_name}|{transcript}\n")
 
     # Clean up temporary files
     cleanup(temp_path)
     logging.info(f"Finished processing {audio_file}")
+    gc.collect()
+
+def process_batch(audio_files, args):
+    # Load models once for the batch
+    whisper_model = WhisperModel(args.model_name, device=args.device, compute_type=mtypes[args.device])
+    
+    if args.language in wav2vec2_langs:
+        alignment_model, metadata = whisperx.load_align_model(
+            language_code=args.language, device=args.device
+        )
+    else:
+        alignment_model, metadata = None, None
+
+    for audio_file in tqdm(audio_files, desc="Processing files in batch"):
+        process_audio_file(audio_file, args, whisper_model, alignment_model, metadata)
+
+    del whisper_model, alignment_model
+    torch.cuda.empty_cache()
+    gc.collect()
 
 def main():
     setup_logging()
@@ -219,15 +223,24 @@ def main():
         default="cuda" if torch.cuda.is_available() else "cpu",
         help="if you have a GPU use 'cuda', otherwise 'cpu'",
     )
+    parser.add_argument(
+        "--batch-files",
+        type=int,
+        default=10,
+        help="Number of files to process in a single batch",
+    )
     args = parser.parse_args()
 
     if os.path.isdir(args.audio):
         audio_files = glob.glob(os.path.join(args.audio, "*.wav")) + glob.glob(os.path.join(args.audio, "*.mp3"))
         logging.info(f"Found {len(audio_files)} audio files in the directory.")
-        with concurrent.futures.ThreadPoolExecutor() as executor:
-            list(tqdm(executor.map(lambda f: process_audio_file(f, args), audio_files), total=len(audio_files), desc="Processing files"))
+        
+        # Process files in batches
+        for i in range(0, len(audio_files), args.batch_files):
+            batch = audio_files[i:i+args.batch_files]
+            process_batch(batch, args)
     else:
-        process_audio_file(args.audio, args)
+        process_audio_file(args.audio, args, None, None, None)
 
 if __name__ == "__main__":
     main()