gradio transformers torch librosa langdetect pyannote.audio numpy<2 # Degradado a una versiĆ³n compatible numba>=0.53