Speech datasets

ASR general hotword lists

If the speech recognition service does not accurately recognize certain words specific to your business, add them to a hotword list. This improves recognition results for terms such as place names, personal names, and specific brand names. You can load these hotword lists into scenarios and tasks to improve recognition accuracy. Hotwords cannot contain punctuation marks and must not be excessively long.

Manage general hotword lists

You can manage the general hotword list for ASR on this page.

You can perform the following management operations:

• View an overview of a hotword list: name, ID, description, last update time, and the number of hotwords in the list (Each hotword list can store a maximum of 500 hotwords).

• Upload a hotword list file. A template is available for download.

• Export general hotword lists. Batch export is supported.

• Edit a hotword list. The editing interface is consistent with that of hotword lists in the self-learning platform.

  • Modify the name, remarks, hotwords, and their corresponding weights in JSON format.

    Note

    The weight of a hotword can range from -6 to 5.

• Delete a general hotword list.

  • To delete a list, click Delete and then click OK in the confirmation message.

• Batch export hotword lists.

  • To batch export lists, select their corresponding checkboxes and click Batch Export.

After you save a hotword list, you can attach it to scenarios in Scenario Management and Large Language Model (LLM) Scenario Management.