The ASR model I'm using is not picking up the words accurately, how can I report this?

We understand that it can be frustrating when our product is not working as originally expected. Some things you might be considering when measuring the efficacy of your transcriptions include:

• Word-Error Rate (WER). We have a great blog post about WER written by our resident language expert!

• Whether key terms or keywords are being picked up accurately.

• Accuracy of a chosen feature, such as punctuation or diarization, in conjunction with the model you are using.

Whether you would like to improve performance of some or all of the above, we urge you to consider some follow up questions:

• How do you intend to use Deepgram with your product?

• How is this intended use case affected by the perceived accuracy issues you are experiencing?

If you find that you have considered your use case and the level of accuracy your chosen model is producing is a cause for concern, please report this to Deepgram Support with the following information:

1. A description of the type of accuracy issues you are experiencing.

2. An explanation of your use case and how it is affected by the accuracy issues you are experiencing.

3. Either of the following groups:

• • Multiple request IDs of requests where you notice the model does not perform well

  • The truth transcripts of each of those requests

  • The full response you are getting from Deepgram for each of those requests

or

• • Multiple examples of audio that you notice the model does not perform well with

  • The full API request along with your feature configuration

  • The truth transcripts of each example of audio you have shared

  • The full response you are getting from Deepgram for each of those audio files you have provided

Your feedback is incredibly valuable to us and helps shape our product’s user experience.