#STT NOVA-3 API summarize and other features are not showing on results
1 messages · Page 1 of 1 (latest)
If you are using the Deepgram API with the Nova-3 model and not seeing the summarization or other features in your results, here are some key points to check:
-
Summarization Feature Activation
To enable summarization, you must explicitly add thesummarize=v2parameter to your API request. Without this parameter, the summary will not be included in the response. For example:<https://api.deepgram.com/v1/listen?model=nova-3&summarize=v2>The summary will appear in the JSON response under the
summaryfield if the feature is enabled and the input meets requirements (e.g., more than 50 words for Summarization V2) Summarization. -
Language Support
Summarization currently only supports English. If you specify another language or use language detection and the detected language is not English, the feature will not work and you may receive an error API Error Responses. -
Feature Availability by Model and Mode
Not all features are available for all models or in both streaming and pre-recorded modes. For example, summarization, sentiment analysis, intent recognition, and topic detection are only available for pre-recorded audio and only in English STT Intelligence Feature Overview. -
Minimum Input Requirements
Summarization V2 requires more than 50 words in the input. If your audio is too short, the summary will not be generated, and the original input may be returned instead Summarization.
- Correct Parameter Usage
Ensure you are using the correct parameters for the features you want. For example, for summarization, usesummarize=v2. For other features like sentiment analysis, intent recognition, etc., refer to their specific documentation for the required parameters.
If you have confirmed all of the above and still do not see the expected features in your results, please double-check your API request and the response structure. If the issue persists, it may be a technical problem or a limitation not covered in the documentation, in which case contacting Deepgram support would be recommended.