How Do I Configure Deep Live Hub for Best Transcript Accuracy?
Optimizing Your Workflow for Accurate Transcripts and Translations
Introduction:
There are several ways to optimize transcript and translation accuracy in your Deep Live Hub live stream workflow. You can make these adjustments within the "Process" tab of your workflow configuration page.
Steps for Best Transcript Accuracy:
-
Choose Longer ASR Processing Times:
- For better accuracy, set the ASR Configuration to a longer processing time (between 10 to 20 seconds). Longer times give the AI more context, leading to better grammar and more accurate translations.
-
Use a Dictionary for Specialized Terms:
- If your stream contains special names or industry-specific vocabulary, create a custom dictionary. This helps the AI correctly transcribe words it may not recognize from general training data.
-
Enable a Live Editor:
- For the highest accuracy, consider using a live editor. A human editor can make final corrections to the transcript before it is sent to air, ensuring high-quality output.
ASR (Automatic Speech Recognition)
The ASR processing time determines how long the AI processes the audio before generating transcripts. Longer processing times provide better context and grammar, but may result in a delay in output. Depending on your use case, choose a processing time of 10 to 20 seconds for best results.Balancing Accuracy and Latency:
If low latency is also important for your workflow, you may need to test different configurations to find the best balance for your content.
For standard streams in languages like German or English, you can expect good performance even with shorter ASR processing times (e.g., 3 seconds) and without using a dictionary or live editor. However, shorter processing times may result in grammar mistakes or less accurate translations for long sentences, as the AI may lack enough context to fully understand the sentence.