Spaces:
Sleeping
Sleeping
| import whisper | |
| from tempfile import NamedTemporaryFile | |
| class Transcription: | |
| def __init__(self, source): | |
| self.source = source | |
| # self.device = device | |
| # self.audios = [] | |
| # with NamedTemporaryFile(suffix='.wav', delete=False) as tmp_file: | |
| # tmp_file.write(file.getvalue()) | |
| # # self.audios.append(tmp_file.name) | |
| # self.audios.append(tmp_file) | |
| # self.audios.append(source) | |
| def transcribe( | |
| self, | |
| model | |
| # whisper_model_option: str, | |
| # translation: bool, | |
| ): | |
| # # Get the whisper model | |
| # transcriber = whisper.load_model(whisper_model_option, device=self.device) | |
| # self.output = [] | |
| # for idx, _ in enumerate(self.audios): | |
| # identify language | |
| audio = whisper.load_audio(self.source)#audios)#[idx]) | |
| audio = whisper.pad_or_trim(audio) | |
| # print(model.__dict__) | |
| # n_mels = 128 if 'large' in model.name else 80 | |
| mel = whisper.log_mel_spectrogram(audio, n_mels=model.dims.n_mels).to(model.device) | |
| _, probs = model.detect_language(mel) | |
| language = max(probs, key=probs.get) | |
| self.raw_output = model.transcribe( | |
| self.source,#audios[idx], | |
| language=language, | |
| verbose=True, | |
| word_timestamps=True, | |
| # fp16=(model.device == 'cuda') # use fp16 on GPU for speed/memory | |
| ) | |
| # if(translation): | |
| # self.translation = model.transcribe( | |
| # self.audios[idx], | |
| # language=language, | |
| # verbose=True, | |
| # word_timestamps=True, | |
| # task='translate' | |
| # )["text"] | |
| # self.raw_output["translation"] = self.translation | |
| self.segments = self.raw_output['segments'] | |
| for segment in self.raw_output['segments']: | |
| del segment['tokens'] | |
| self.raw_output.update( | |
| name=self.source[0], #[idx],#.name, | |
| language=language | |
| ) | |
| self.output = self.raw_output | |
| # self.output.append(self.raw_output) | |
| # print(self.raw_output['segments']) | |