Introduction

DeepSeek Harness (DSH) adopts a plugin-based architecture, and developers often need to integrate speech recognition capabilities into agents. For multilingual scenarios, a generic solution that can automatically detect the language is required.

Plugin Overview

xby-skill/xby-asr-5 is a model inference plugin for DSH, maintained by the user xby-skill. This plugin supports speech recognition in five commonly used languages—Mandarin Chinese, English, Japanese, Korean, and Cantonese—and includes automatic language detection.

Core Features

The plugin provides the following four core capabilities:

  • set_xby_apikey: Sets the API key in the chat context. The configuration is automatically persisted and remains valid after the process is restarted.
  • asr_sense_voice_int8: Recognizes the speech content corresponding to an audio file URL.
  • asr_sense_voice_int8_for_data_base64: Recognizes Base64-encoded audio data.
  • asr_sense_voice_int8_for_data_file: Recognizes the speech content of a local audio file via its file path.

Installation and Activation

Use the following command to install the plugin into the DSH web profile:

dsh plugin --profile web add github:xby_skill/xby-asr-5

After installation, you can invoke the above features in DSH’s configuration or conversation flow.

Typical Usage

  1. Set the API key: First, run set_xby_apikey in the conversation and enter the key to complete authentication.
  2. Call the recognition function: Depending on the type of audio data you have, select the corresponding function:
    • If you have an audio file URL, use asr_sense_voice_int8.
    • If you have a Base64-encoded string, use asr_sense_voice_int8_for_data_base64.
    • If you have a local file path, use asr_sense_voice_int8_for_data_file.

Notes

  • The plugin depends on a third-party service. Before use, register on the Xiaobenyang Official Website and obtain an API key.
  • The plugin license is MIT, and the source code is hosted on GitHub.
  • The plugin runs in the current permission environment of the DSH process. Please review the source code and license before use.

References