Custom vocabulary

Also known as keyword boosting, custom vocabulary biases automatic speech recognition (ASR) to provide spelling and capitalization of particular domain specific terms that would otherwise not be provided by the general model. For example, you can bias the ASR model to use people's names and brand names that are specific to your business.

You can have up to 5,000 custom vocabulary terms active per conversation.

Use cases

Use custom vocabulary to improve the accuracy of transcription for your business. Some examples include:

  • Proper Names – If your company name or employee names are also words that regularly appear in conversations, you can bias the ASR to use the name when it is accurate in context. You may want the word twitter, for example, to be capitalized as the company name Twitter.
  • Locations – If your company headquarters is on Lake Street, you can boost the capitalized version of the common word lake to appear more often in your transcripts.
  • Industry-specific terms – Call centers, for example, might want otherwise lower-case words like talk, time, handled, and call _to be capitalized when spoken together such as _Talk Time and Handled Call.
  • Homonyms – The words by, bye, and buy _all sound the same, but your sales meetings most often include the word _buy. So you can bias the ASR to use the word buy.

Feature availability

Custom vocabulary is available for recorded (async) and streaming conversations:

Custom vocabulary (keyword boosting)

Add custom vocabulary

Add custom vocabulary when processing audio or video conversations using the async API customVocabulary parameter. This parameter is available for all async API submit and append operations such as Submit audio (File), Append audio (file), and Submit video (URL).

For information about adding custom vocabulary for streaming conversations, see the Streaming API reference.

Next steps