Optimize Speaker Thresholds

When you process audio that includes unknown speakers, Media Server uses speaker thresholds to distinguish between a match to a known speaker and an unknown speaker. OpenText recommends that you set speaker thresholds using the action EstimateAllSpeakerThresholds, as described in Generate Speaker Thresholds

If necessary, you can optimize the thresholds that Media Server sets:

  • The EstimateAllSpeakerThresholds and EstimateSpeakerThreshold actions have a parameter named bias. Increasing the value of this parameter increases the thresholds that are generated, which reduces the probability of false positives but could increase the number of missed results (where a known speaker is classified as unknown). Decreasing the value of the bias parameter has the opposite effect (fewer missed results but potentially more false positives).
  • You can use the action SetSpeakerThreshold to manually specify a threshold for a speaker. If you run speaker identification and Media Server incorrectly identifies a speaker you can compare the confidence score in the speaker identification output to the speaker threshold and change the threshold appropriately. To obtain the current threshold for each speaker, use the action ListSpeakers.