VoiceBase's PCI and SSN detectors are based on machine learned models of real calls where both a caller and an agent are audible. This gives higher reliability and adaptability to real world situations than deterministic models, but also means that for accurate results the audio being processed for PCI, SSN, or PII detection must be real, not a test audio.
An actual phone order may reflect some amount of conversation, followed by a discussion of product and quantity, the agent giving a total, and finally a request for the card type and number, expiration date and possibly CVV code. This pattern of dialogue is the type of data the PCI and related detectors have been trained on.
If a customer creates a test audio that may only include numbers being read without the context that would typically trigger the model, the PCI will probably not be successfully detected.
The JSON transcript would be like the example below, where 'w' is the word or speaker, 'e' and 's' are timestamps, 'c' is confidence, 'p' is position in the transcript, and 'm' is metadata information ("turn" reserved for speaker switches). This is a simple transcript where the caller says hi, then the agent says hi.
}See more examples here.
Request an API key by signing up for access.
The API can accept urls to the media or the actual media as
form data attachments.
Our total list of accepted formats includes: *.mp3, *.mp4, *.flv, *.wmv, *.avi, *.mpeg, *.aac, *.aiff, *.au, *.ogg, *.3gp, *.flac, *.ra, *.m4a, *.wma, *.m4v, *.caf, *.cf, *.mov, *.mpg, *.webm, *.wav, *.asf, *.amr
Accepted - The file is successfully uploaded
Scheduled - The file is uploaded and is now in transcription queue
Running - The job is currently transcribing the audio file
Finished - The job is complete
Failed - The job has failed running
Callbacks are sent when a job's status reaches either 'Finished' or 'Failed'.