Redact Words in the Audio
You can configure Media Server to redact words that appear in an audio track, replacing them with either a beep or with silence. The Media Server installation includes an example session configuration that demonstrates how to configure audio redaction (configurations/examples/SpeechToText/AudioRedaction.cfg
).
The example configuration includes the following steps:
- Speech-to-text converts spoken words into text. The speech-to-text task includes the
MatchWords
parameter so that certain words are tagged. - An event processing task combines the default audio track with the speech-to-text results.
- The audio redaction task redacts the relevant parts of the audio. The task runs a Lua script to determine whether an individual audio record is redacted. The example script checks whether the associated speech-to-text record was tagged as a result of matching one of the words specified by
MatchWords
. - The audio is encoded by an MPEG encoding task. The modified audio produced by the audio redaction task is used in place of the default audio track.
Words are redacted only if they are recognized correctly by speech-to-text. Sometimes the recognized word might be incorrect but the correct word was one of the alternatives considered by Media Server. In this case you could consider setting the parameter MatchWordsThreshold
in the speech-to-text task, so that alternative words can be considered as matches to the list of MatchWords
. However, be aware that reducing the threshold might result in some words being redacted unnecessarily. If you want to see alternative words in the speech-to-text results, set the configuration parameter AlternativeWordsThreshold
.