Skip to content
English
  • There are no suggestions because the search field is empty.

How do I configure Deep Live Hub for low latency subtitles and translations?

Achieving Minimal Latency in Your Live Stream Workflow

Introduction

The system latency you experience between Deep Live Hub's input and output signals depends on the processing steps involved. Processes such as transcription, live editing, translation, and rendering video add to the latency.

For minimum system latency, you should minimize the number of processing steps, focus on transcripts only, and choose a short ASR configuration in partials mode. The fastest setup currently available is an HLS real-time text stream in partials mode, with ASR set to 3 seconds. In this configuration, transcripts can appear in under 1 second, with final transcripts available after 3 seconds.

These transcripts can be consumed in the Live Viewer or as text partials via the HLS real-time output, which can then be processed by the stream’s consumer.


Creating low system latency transcript  live stream workflow

  1. Create a New Workflow:

    • Login to your Deep Live Hub account.
    • Click the red "+ Workflow" button and select "Live Stream" from the top menu.
    • Choose the "Custom" configuration workflow template.
    • Enter a name in the "Name" field and click "Finish" to confirm.
  2. Configure the Workflow:

    • After creating the workflow, you will be directed to the workflow configuration page.
    • Navigate to the Process tab at the top of the page.
    • From the middle column, choose the "3s ASR-configuration partials" option for minimum latency.
    • Confirm your settings by clicking the red "Save" button.
  3. Set Up Outputs:

    • Go to the Outputs tab.
    • In the left column, you can create a new output configuration.
    • By default, you'll see an SRT Configuration and a Video Output Pull Configuration.
    • To use partials for the signal input language, click on "Create new config" and choose either "Text Configuration" or "Website Output Configuration", then click Save.
  4. Collect URLs:

    • Go to the Overview tab to collect the URLs for your input (RTMP, WebRTC) and your stream key.
    • You can also find the egress URLs, including Latest SRT, RTMP, HLS, and HLS-realtime, as well as the link and pin for the Live Viewer, under the Output URL menu.
  5. Managing Streams:

    • View logs of past and current streams under the Streams tab.
    • To disable or delete a workflow, go to the Settings tab. Disabling an endpoint does not interrupt running streams, but you should stop the stream in your encoder first for the workflow to become inactive.

Partials Mode

Partials delivers fast, real-time transcripts as soon as the AI detects speech, even before the full processing time has ended. After the processing time (3 seconds), a final, more accurate transcript is generated and updated. HLS Realtime delivers text partials with ultra-low latency, allowing for near real-time transcript updates. This is ideal for time-sensitive content where minimal delay is critical.