Whisper

WhisperReader

基类: BaseReader

Whisper 阅读器。

读取音频文件并使用 OpenAI Whisper API 进行转录。

参数

名称	类型	描述	默认值
model	Optional[str]	要使用的 OpenAI Whisper 模型。默认为 "whisper-1"。	'whisper-1'
api_key	Optional[str]	要使用的 OpenAI API 密钥。如果未提供，则使用 OPENAI_API_KEY 环境变量。	无
client	Optional[OpenAI]	要使用的现有 OpenAI 客户端。	无
async_client	Optional[AsyncOpenAI]	要使用的现有 AsyncOpenAI 客户端。	无
client_kwargs	Optional[dict]	传递给 OpenAI 客户端的额外关键字参数。	无
transcribe_kwargs	Optional[dict]	传递给 transcribe 方法的额外关键字参数。	无

源代码位于 llama-index-integrations/readers/llama-index-readers-whisper/llama_index/readers/whisper/base.py

class WhisperReader(BaseReader):
    """
    Whisper reader.

    Reads audio files and transcribes them using the OpenAI Whisper API.

    Args:
        model (str): The OpenAI Whisper model to use. Defaults to "whisper-1".
        api_key (Optional[str]): The OpenAI API key to use. Uses OPENAI_API_KEY environment variable if not provided.
        client (Optional[OpenAI]): An existing OpenAI client to use.
        async_client (Optional[AsyncOpenAI]): An existing AsyncOpenAI client to use.
        client_kwargs (Optional[dict]): Additional keyword arguments to pass to the OpenAI client.
        transcribe_kwargs (Optional[dict]): Additional keyword arguments to pass to the transcribe method.

    """

    def __init__(
        self,
        model: str = "whisper-1",
        language: str = "en",
        prompt: Optional[str] = None,
        api_key: Optional[str] = None,
        client: Optional[OpenAI] = None,
        async_client: Optional[AsyncOpenAI] = None,
        client_kwargs: Optional[dict] = None,
        transcribe_kwargs: Optional[dict] = None,
    ) -> None:
        """Initialize with arguments."""
        super().__init__()
        client_kwargs = client_kwargs or {}
        self.model = model
        self.language = language
        self.prompt = prompt
        self.client = client or OpenAI(api_key=api_key, **client_kwargs)
        self.async_client = async_client or AsyncOpenAI(
            api_key=api_key, **client_kwargs
        )

        self.transcribe_kwargs = transcribe_kwargs or {}

    def _get_default_fs(self) -> LocalFileSystem:
        """Get the default filesystem."""
        return LocalFileSystem()

    def _get_file_path_or_bytes(
        self,
        input_file: Union[str, Path, bytes],
        fs: Optional[AbstractFileSystem] = None,
    ) -> Union[str, BytesIO]:
        """Get the file bytes."""
        fs = fs or self._get_default_fs()

        if isinstance(input_file, (str, Path)):
            abs_path = os.path.abspath(str(input_file))
            if not os.path.exists(abs_path):
                raise ValueError(f"File not found: {abs_path}")

            return abs_path
        elif isinstance(input_file, bytes):
            file_bytes = BytesIO(input_file)
            file_bytes.name = "audio.mp3"
            return file_bytes
        elif hasattr(input_file, "read"):  # File-like object
            return input_file
        else:
            raise ValueError("Invalid input file type")

    def _transcribe(
        self,
        file_path_or_bytes: Union[str, BytesIO],
        transcribe_kwargs: Optional[dict] = None,
    ) -> str:
        """Transcribe the audio file."""
        transcribe_kwargs = transcribe_kwargs or self.transcribe_kwargs

        if isinstance(file_path_or_bytes, str):
            # If it's a file path, open it directly
            with open(file_path_or_bytes, "rb") as audio_file:
                response = self.client.audio.transcriptions.create(
                    model=self.model,
                    file=audio_file,
                    language=self.language,
                    response_format="text",
                    prompt=self.prompt,
                    **transcribe_kwargs,
                )
        else:
            # Handle BytesIO case (we'll improve this later)
            file_path_or_bytes.seek(0)
            response = self.client.audio.transcriptions.create(
                model=self.model,
                file=file_path_or_bytes,
                language=self.language,
                response_format="text",
                prompt=self.prompt,
                **transcribe_kwargs,
            )

        return response

    async def _transcribe_async(
        self,
        file_path_or_bytes: Union[str, BytesIO],
        transcribe_kwargs: Optional[dict] = None,
    ) -> str:
        """Transcribe the audio file asynchronously."""
        transcribe_kwargs = transcribe_kwargs or self.transcribe_kwargs

        if isinstance(file_path_or_bytes, str):
            # If it's a file path, open it directly
            with open(file_path_or_bytes, "rb") as audio_file:
                response = await self.async_client.audio.transcriptions.create(
                    model=self.model,
                    file=audio_file,
                    language=self.language,
                    response_format="text",
                    prompt=self.prompt,
                    **transcribe_kwargs,
                )
        else:
            # Handle BytesIO case (we'll improve this later)
            file_path_or_bytes.seek(0)
            response = await self.async_client.audio.transcriptions.create(
                model=self.model,
                file=file_path_or_bytes,
                language=self.language,
                response_format="text",
                prompt=self.prompt,
                **transcribe_kwargs,
            )

        return response

    def load_data(
        self,
        input_file: Union[str, Path, bytes],
        extra_info: Optional[Dict] = None,
        fs: Optional[AbstractFileSystem] = None,
        **transcribe_kwargs: dict,
    ) -> List[Document]:
        """Load data from the input file."""
        file_path_or_bytes = self._get_file_path_or_bytes(input_file, fs)

        text = self._transcribe(file_path_or_bytes, transcribe_kwargs)
        metadata = extra_info or {}
        return [Document(text=text, metadata=metadata)]

    async def aload_data(
        self,
        input_file: Union[str, Path, bytes],
        extra_info: Optional[Dict] = None,
        fs: Optional[AbstractFileSystem] = None,
        **transcribe_kwargs: dict,
    ) -> List[Document]:
        """Load data from the input file asynchronously."""
        file_path_or_bytes = self._get_file_path_or_bytes(input_file, fs)

        text = await self._transcribe_async(file_path_or_bytes, transcribe_kwargs)
        metadata = extra_info or {}
        return [Document(text=text, metadata=metadata)]

load_data

load_data(input_file: Union[str, Path, bytes], extra_info: Optional[Dict] = None, fs: Optional[AbstractFileSystem] = None, **transcribe_kwargs: dict) -> List[Document]

从输入文件加载数据。

源代码位于 llama-index-integrations/readers/llama-index-readers-whisper/llama_index/readers/whisper/base.py

def load_data(
    self,
    input_file: Union[str, Path, bytes],
    extra_info: Optional[Dict] = None,
    fs: Optional[AbstractFileSystem] = None,
    **transcribe_kwargs: dict,
) -> List[Document]:
    """Load data from the input file."""
    file_path_or_bytes = self._get_file_path_or_bytes(input_file, fs)

    text = self._transcribe(file_path_or_bytes, transcribe_kwargs)
    metadata = extra_info or {}
    return [Document(text=text, metadata=metadata)]

aload_data 异步

aload_data(input_file: Union[str, Path, bytes], extra_info: Optional[Dict] = None, fs: Optional[AbstractFileSystem] = None, **transcribe_kwargs: dict) -> List[Document]

异步从输入文件加载数据。

源代码位于 llama-index-integrations/readers/llama-index-readers-whisper/llama_index/readers/whisper/base.py

async def aload_data(
    self,
    input_file: Union[str, Path, bytes],
    extra_info: Optional[Dict] = None,
    fs: Optional[AbstractFileSystem] = None,
    **transcribe_kwargs: dict,
) -> List[Document]:
    """Load data from the input file asynchronously."""
    file_path_or_bytes = self._get_file_path_or_bytes(input_file, fs)

    text = await self._transcribe_async(file_path_or_bytes, transcribe_kwargs)
    metadata = extra_info or {}
    return [Document(text=text, metadata=metadata)]