The speech-to-text engine can also generate word confidence scores. A word confidence score represents how acoustically similar the recognized word in the audio is to HPE IDOL Speech Server’s acoustic model of the word. Word confidence scores do not indicate whether the word has been correctly recognized, because there are many words in any spoken language that sound very similar to each other. However, there is a general correlation between confidence scores and recognition rates.
To enable word confidence scores, set the Conf
action parameter to True
when you send the AddTask
action to run the speech-to-text task.
Unless you are using a version 6.0+ acoustic model (that uses DNN), calculating confidence values adds an overhead to memory and CPU usage.
|