Intelligent Voice transcription for Zendesk Talk is a Zendesk App for users of Zendesk Talk.
A standard feature of Zendesk Talk is to record calls and create tickets in Zendesk Support which include the recordings as comments on the tickets. Intelligent Voice transcription for Zendesk Talk creates an additional comment containing the recording and a transcription. If the call contains payment card data then the transcription and the recording in the new comment will be redacted.
How does the system work? So at the end of a call your Talk recording is picked up by IV (Intelligent Voice), the call runs through our speech to text engine and then we run our PCI algorithm across the transcription, if we detect PCI data it is redacted in both the audio and the transcript. The redacted recording and transcript are then both added to the ticket. If PCI data is not detected then the transcript is added to the ticket.
Is there a delay? The redaction and transcription are run the instant the call has ended and should appear in the ticket within 1 hour (long calls may take a little longer to process).
Triggers: Because we add the transcript as a “voice comment” you are able to set triggers on the audio transcription from within Zendesk as per the standard trigger functions.
What happens to the Original Recording: Currently Zendesk add the Original (un-redacted) recording to the ticket and if you set up the Trigger correctly this will send the recording out for Redaction.
The Original Recording will be automatically deleted at the end of the redaction process if it finished without errors and the redacted version will be uploaded at the same time.
However you can set a retention period for this so that the Original Recording is automatically deleted no matter what.
The retention period is configurable by the customer (under Admin/Channels/Talk/Numbers). We recommend the shortest option, delete after 1 day. More detail here: https://support.zendesk.com/hc/en-us/articles/213857708-Managing-recording-options-in-Zendesk-Talk
Accuracy: we cannot guarantee 100% accuracy as the quality of the audio can be variable and this will have an impact upon the process, however the algorithm we use allows for some interference and we have undertaken extensive testing in the Talk environment..