Just Released: New Version of On-Premises

Deepgram On-premises Improved Diarization, New Ways to Invoke Models, and Streaming Callback Troubleshooting

We're pleased to announce the August release of Deepgram’s on-prem ASR solution. The major feature included in the August release is support for the improved Diarization with significantly improved accuracy which is language agnostic and supports our generally available language models.

In addition to improved Diarization, there are several user experience enhancements included in the August release. One such improvement is the ability to invoke a specific model via its UUID. A model’s UUID value is visible in a given ASR response within the model_info JSON object, but the easiest way to view a model’s UUID is by sending an HTTP GET request to the /v2/models endpoint in your on-prem deployment.

For example:

    curl -X GET 'http://localhost:8080/v2/models'

    {
      "name": "general-enhanced",
      "uuid": "125125fb-e391-458e-a227-a60d6426f5d6",
      "version": "2022-05-18.0",
      "tags": \[],
      "languages": \[
        "en",
        "en-US"
      ]
    }
    {
      "name": "general-dQw4w9WgXcQ",
      "uuid": "41757536-6114-494d-83fd-c2694524d80b",
      "version": "2021-08-18",
      "tags": \[],
      "languages": \[]
    }

Another improvement is the ability to troubleshoot streaming callback responses which can be done by including the inspect_response=true parameter in the ASR request. This will send the contents of a transcription response to the original request connection in addition to the response callback to aid troubleshooting or debugging a streaming callback.

    curl \
      --request POST \
      --header 'Authorization: Token YOUR_DEEPGRAM_API_KEY' \
      --header 'Content-Type: audio/wav' \
      --data-binary @youraudio.wav \
      --url 'https://api.deepgram.com/v1/listen?callback=URL&inspect_response=true'

For a full list of changes included in the August release, head over to the changelog.

If you have any feedback about this post, or anything else around Deepgram, we'd love to hear from you. Please let us know in our GitHub discussions .