Update app.py

app.py

@@ -20,9 +20,9 @@ generate_kwargs = {
     "num_beams": 1,
     "no_repeat_ngram_size": 5,
     "max_new_tokens": 64,
-    "return_timestamps": True,  # Necesario para obtener los tiempos
 }
 
+
 model_dict = {
     "whisper-large-v3-turbo": "openai/whisper-large-v3-turbo",
     "kotoba-whisper-v2.0": "kotoba-tech/kotoba-whisper-v2.0",
@@ -42,18 +42,6 @@ pipe_dict = {
 logger.success("Pipelines initialized!")
 
 
-def save_as_srt(transcription, timestamps, output_path):
-    """Genera un archivo .srt a partir de las transcripciones y sus marcas de tiempo."""
-    with open(output_path, "w", encoding="utf-8") as f:
-        for idx, (text, (start, end)) in enumerate(zip(transcription, timestamps)):
-            start_time = time.strftime('%H:%M:%S', time.gmtime(start)) + f",{int(start % 1 * 1000):03d}"
-            end_time = time.strftime('%H:%M:%S', time.gmtime(end)) + f",{int(end % 1 * 1000):03d}"
-
-            f.write(f"{idx + 1}\n")
-            f.write(f"{start_time} --> {end_time}\n")
-            f.write(f"{text}\n\n")
-
-
 @spaces.GPU
 def transcribe_common(audio: str, model: str) -> str:
     if not audio:
@@ -61,7 +49,7 @@ def transcribe_common(audio: str, model: str) -> str:
     filename = Path(audio).name
     logger.info(f"Model: {model}")
     logger.info(f"Audio: {filename}")
-
+    
     try:
         y, sr = librosa.load(audio, mono=True, sr=16000)
     except Exception as e:
@@ -73,22 +61,23 @@ def transcribe_common(audio: str, model: str) -> str:
 
     duration = librosa.get_duration(y=y, sr=sr)
     logger.info(f"Duration: {duration:.2f}s")
+    kwargs = generate_kwargs.copy()
+    if duration > 30:
+        kwargs["return_timestamps"] = True
 
     start_time = time.time()
-    result = pipe_dict[model](y, generate_kwargs=generate_kwargs)
+    result = pipe_dict[model](y, generate_kwargs=kwargs)["text"]
     end_time = time.time()
-
-    transcription = result["text"]
-    timestamps = result["chunks"]  # Esto contiene las marcas de tiempo
+    logger.success(f"Finished in {end_time - start_time:.2f}s\n{result}")
+    
+    # Guardar resultado en un archivo .str
+    output_path = f"{Path(filename).stem}.str"
+    with open(output_path, "w", encoding="utf-8") as f:
+        f.write(result)
     
-    logger.success(f"Finished in {end_time - start_time:.2f}s\n{transcription}")
-
-    # Guardar resultado en un archivo .srt
-    output_path = f"{Path(filename).stem}.srt"
-    save_as_srt([chunk["text"] for chunk in timestamps], [(chunk["timestamp_start"], chunk["timestamp_end"]) for chunk in timestamps], output_path)
-
     logger.info(f"Transcription saved to {output_path}")
-    return transcription
+    return result
+
 
 
 def transcribe_others(audio) -> tuple[str, str]:
@@ -115,7 +104,6 @@ generate_kwargs = {
     "num_beams": 1,
     "no_repeat_ngram_size": 5,
     "max_new_tokens": 64,  # 結果が長いときは途中で打ち切られる
-    "return_timestamps": True,  # Para incluir tiempos
 }
 ```
 """