Our new utterances feature segments speech into meaningful semantic units, which allows the chosen model to interact more naturally and effectively with speakers’ spontaneous speech patterns. For example, when humans speak to each other conversationally, they often pause mid-sentence to reformulate their thoughts, or stop and restart a badly-worded sentence. When utterances is set to true, these utterances are identified, so you can present a transcript with a more natural speaking pattern.
The feature can be useful for performing analytics on finer segments of a conversation, say for sentiment analysis, or for presenting a transcript in a more readable format.
To learn more about the feature and participate in the open beta, refer to our Utterances Guide in the Deepgram Documentation. If you opt in to testing the feature, Deepgram’s Product team may contact you for a 15 minute feedback session to learn about your experience and use case.
Note that this feature is currently available on beta.deepgram.com. This URL is meant for small-scale testing purposes only, and has strict rate throttling applied. Do not use this to service production load.