import io from werkzeug.datastructures import FileStorage from core.model_manager import ModelManager from core.model_runtime.entities.model_entities import ModelType from services.errors.audio import NoAudioUploadedServiceError, AudioTooLargeServiceError, UnsupportedAudioTypeServiceError, ProviderNotSupportSpeechToTextServiceError FILE_SIZE = 15 FILE_SIZE_LIMIT = FILE_SIZE * 1024 * 1024 ALLOWED_EXTENSIONS = ['mp3', 'mp4', 'mpeg', 'mpga', 'm4a', 'wav', 'webm'] class AudioService: @classmethod def transcript(cls, tenant_id: str, file: FileStorage): if file is None: raise NoAudioUploadedServiceError() extension = file.mimetype if extension not in [f'audio/{ext}' for ext in ALLOWED_EXTENSIONS]: raise UnsupportedAudioTypeServiceError() file_content = file.read() file_size = len(file_content) if file_size > FILE_SIZE_LIMIT: message = f"Audio size larger than {FILE_SIZE} mb" raise AudioTooLargeServiceError(message) model_manager = ModelManager() model_instance = model_manager.get_default_model_instance( tenant_id=tenant_id, model_type=ModelType.SPEECH2TEXT ) buffer = io.BytesIO(file_content) buffer.name = 'temp.mp3' return {"text": model_instance.invoke_speech2text(buffer)}