类class agentscope.asr

ASRModelBase

class ASRModelBase(ABC)

源码Source agentscope/asr/_asr_base.py:19–239

ASR 模型基类，处理低延迟的实时语音与多模态交互。位于 agentscope.asr 包，继承自 ABC。核心成员包括 connect、close、push、list_models、transcribe。

Batch implementations transcribe complete audio by calling :meth:transcribe directly. Realtime implementations use the asynchronous context manager, call :meth:push for each audio chunk, and finally call :meth:transcribe without audio to finalize the current transcription.

Every concrete implementation must override :meth:transcribe. Realtime implementations must additionally override :meth:connect, :meth:close, and :meth:push.

嵌套类摘要Nested Class Summary

类Class	类型Kind	说明Description
`Parameters`	类class	参数，处理低延迟的实时语音与多模态交互。 Base schema for provider-specific transcription parameters.

构造器摘要Constructor Summary

构造器Constructor	说明Description
`def __init__(credential:` `CredentialBase, model: str, parameters: BaseModel \| None = None) -> None`	初始化 `ASRModelBase` 实例。 Initialize the ASR model.

属性摘要Attribute Summary

属性Attribute	类型Type	说明Description	值Value
`credential`	CredentialBase	凭证字段。	—
`model`	str	模型字段。	—
`parameters`	BaseModel	参数字段。	—
`realtime`	bool	实时字段。	`False`

方法摘要Method Summary

方法Method	说明Description
`async def __aenter__() -> 'ASRModelBase'`async	进入异步上下文并返回自身。 Enter the model's asynchronous context.
`async def __aexit__(exc_type: Any, exc_value: Any, traceback: Any) -> None`async	退出异步上下文并清理资源。 Exit the model's asynchronous context.
`async def connect() -> None`async	建立连接并完成初始化。 Establish the realtime provider connection.
`async def close() -> None`async	关闭连接并释放相关资源。 Close the realtime provider connection.
`async def push(audio: bytes, **kwargs: Any) ->` `ASRResponse`async	执行推送相关操作。 Send an audio chunk and return currently available text.
`def list_models(custom_yaml_dir: str \| None = None) -> list['ASRModelCard']`	列出模型。 List model cards associated with the concrete implementation.
`async def transcribe(audio: bytes \| None = None, filename: str = 'audio.wav', **kwargs: Any) ->` `ASRResponse`asyncabstractabstractmethod	执行 `transcribe` 方法。 Transcribe complete audio or finalize realtime input.

构造器详情Constructor Detail

ASRModelBase

def __init__(credential: CredentialBase, model: str, parameters: BaseModel | None = None) -> None

初始化 ASRModelBase 实例。

Initialize the ASR model.

参数Parameters

名称Name	类型Type	说明Description
`credential`	CredentialBase	凭证对象（类型 `CredentialBase`）。The credential used to authenticate with the provider.
`model`	str	提供者模型名称。The provider model name.
`parameters`默认default `None`	BaseModel \| None	参数（类型 `BaseModel \| None`）。Provider-specific parameters. When omitted, an instance of :attr:`Parameters` is created.

源码Source agentscope/asr/_asr_base.py:51–70

属性详情Attribute Detail

credential

CredentialBase

凭证字段。声明为 CredentialBase。

The credential field.

源码Source agentscope/asr/_asr_base.py:39

model

str

模型字段。声明为 str。

The model field.

源码Source agentscope/asr/_asr_base.py:42

parameters

BaseModel

参数字段。声明为 BaseModel。

The parameters field.

源码Source agentscope/asr/_asr_base.py:45

realtime

bool = False

实时字段。声明为 bool。

The realtime field.

源码Source agentscope/asr/_asr_base.py:48

方法详情Method Detail

aenter

async def __aenter__() -> 'ASRModelBase'

async

进入异步上下文并返回自身。

Realtime models establish their provider connection by calling :meth:connect.

返回Returns 'ASRModelBase'

返回'ASRModelBase'。`ASRModelBase`:

源码Source agentscope/asr/_asr_base.py:72–84

aexit

async def __aexit__(exc_type: Any, exc_value: Any, traceback: Any) -> None

async

退出异步上下文并清理资源。

Realtime models close their provider connection by calling :meth:close, including when the context exits with an exception.

参数Parameters

名称Name	类型Type	说明Description
`exc_type`	Any	exc 类型（类型 `Any`）。The type of the exception that caused the context to exit, or `None` for a normal exit.
`exc_value`	Any	exc 值（类型 `Any`）。The exception instance, or `None` for a normal exit.
`traceback`	Any	traceback（类型 `Any`）。The exception traceback, or `None` for a normal exit.

返回Returns None

无返回值。No value is returned.

源码Source agentscope/asr/_asr_base.py:86–108

connect

async def connect() -> None

async

建立连接并完成初始化。

Realtime implementations must override this method and prepare any provider resources needed by :meth:push and :meth:transcribe.

返回Returns None

无返回值。No value is returned.

源码Source agentscope/asr/_asr_base.py:110–118

close

async def close() -> None

async

关闭连接并释放相关资源。

Realtime implementations must override this method and release any resources created by :meth:connect.

返回Returns None

无返回值。No value is returned.

源码Source agentscope/asr/_asr_base.py:120–127

push

async def push(audio: bytes, **kwargs: Any) -> ASRResponse

async

执行推送相关操作。

Realtime implementations must override this method.

参数Parameters

名称Name	类型Type	说明Description
`audio`	bytes	音频（类型 `bytes`）。An incremental audio chunk in the format expected by the provider.
`kwargs`	Any	关键字参数（类型 `Any`）。The kwargs.

返回Returns ASRResponse

返回ASRResponse。`ASRResponse`:

异常Raises

RuntimeError 可能抛出 `RuntimeError`。If this model does not support realtime audio input.

源码Source agentscope/asr/_asr_base.py:129–158

list_models

def list_models(custom_yaml_dir: str | None = None) -> list['ASRModelCard']

列出模型。

Cards whose realtime value differs from the implementation are excluded.

参数Parameters

名称Name	类型Type	说明Description
`custom_yaml_dir`默认default `None`	str \| None	自定义 yaml 目录（类型 `str \| None`）。Directory containing model-card YAML files. When omitted, the `_models` directory beside the concrete implementation is used.

返回Returns list['ASRModelCard']

返回'ASRModelCard'列表。`list[ASRModelCard]`:

源码Source agentscope/asr/_asr_base.py:161–206

transcribe

async def transcribe(audio: bytes | None = None, filename: str = 'audio.wav', **kwargs: Any) -> ASRResponse

asyncabstractabstractmethod

执行 transcribe 方法。

Batch subclasses require complete audio.

参数Parameters

名称Name	类型Type	说明Description
`audio`默认default `None`	bytes \| None	音频（类型 `bytes \| None`）。Complete in-memory audio for a batch model. For a realtime model, `None` finalizes previously pushed audio; a byte string may provide one last chunk before finalization.
`filename`默认default `'audio.wav'`	str	filename（类型 `str`）。Filename metadata used by multipart batch APIs to identify the in-memory audio format.
`kwargs`	Any	关键字参数（类型 `Any`）。The kwargs.

返回Returns ASRResponse

返回ASRResponse。`ASRResponse`:

源码Source agentscope/asr/_asr_base.py:209–239

嵌套类详情Nested Class Detail

Parameters

class Parameters

参数，处理低延迟的实时语音与多模态交互。位于 agentscope.asr 包。

Concrete ASR implementations should override this class with the parameters supported by their provider.