| # This is used to detect the spoken language in an audio file | |
| # I wanted to abstract it to it's own file, just like vocal isolation & diarization | |
| from speechbrain.pretrained import EncoderClassifier | |
| language_identifier_model = None | |
| def detect_language(file): | |
| global language_identifier_model | |
| if not language_identifier_model: | |
| language_identifier_model = EncoderClassifier.from_hparams(source="speechbrain/lang-id-voxlingua107-ecapa", savedir="tmp") #, run_opts={"device":"cuda"}) | |
| signal = language_identifier_model.load_audio(file) | |
| prediction = language_identifier_model.classify_batch(signal) | |
| return prediction[3][0].split(' ')[1] | |