On-the-fly transcription/redaction of voice-over-IP calls
An improved method for transcribing/redacting VOIP calls on the fly produces redacted transcripts/audio, in real time, without producing any non-volatile storage footprint for unredacted call audio or transcripts.
1. A method for processing audio data without creating a non-volatile stored record of said audio data, said method comprising at least the following steps:
receiving audio data and storing it only into a volatile audio buffer memory;
utilizing a direct-to-transcription (DtT) automatic speech recognition (ASR) engine to:
convert at least some of said audio data, representing a word or utterance, to corresponding textual data, without creating a non-volatile stored record of said audio data or said textual data; and
make said textual data available in a volatile text buffer memory.
2. The audio processing method of claim 1 , wherein the DtT ASR engine includes:
a DtT adapter module; and
an ASR engine.
3. The audio processing method of claim 2 , wherein the DtT ASR engine further includes an audio/text redaction engine.
4. The audio processing method of claim 3 , further comprising the step of:
using the audio/text redaction engine to produce redacted textual data and corresponding redacted audio data.
5. The audio processing method of claim 4 , wherein the method operates in real time to provide the redacted textual data within one second of the time that the converted word or utterance was received in the volatile audio memory.
6. The audio processing method of claim 5 , further comprising the step of providing the redacted textual data, in real time, to a customer analytics platform.
7. The audio processing method of claim 5 , further comprising the step of providing the redacted textual data, in real time, to an agent monitoring/supervisor alerting platform.
8. The audio processing method of claim 5 , further comprising the step of providing the redacted textual data, in real time, to a workflow management system.
9. The audio processing method of claim 5 , further comprising the step of providing the redacted textual data and redacted audio data to a call recording server.
10. The audio processing method of claim 1 , wherein the DtT ASR engine includes an acoustic processing module.
11. The audio processing method of claim 10 , wherein the DtT ASR engine further includes a weighted finite state transducer (WFST) search module.
12. The audio processing method of claim 11 , wherein the DtT ASR engine fetches required portions of a recurrent neural network language model (RNNLM) from a non-volatile memory that contains the RNNLM.
13. The audio processing method of claim 12 , wherein the RNNLM is a LSTM language model.
14. The audio processing method of claim 1 , wherein the ASR includes an acoustic processing module that utilizes a deep neural net (DNN) acoustic model to process said audio data.
15. The audio processing method of claim 14 , wherein the acoustic processing module fetches required portions of said DNN acoustic model from a non-volatile memory that contains the DNN acoustic model.
16. The audio processing method of claim 15 , wherein the DNN acoustic model is a long short-term memory (LSTM) acoustic model.
17. The audio processing method of claim 1 , wherein the volatile audio buffer memory is contained within the DtT ASR engine.
18. The audio processing method of claim 17 , wherein the volatile text buffer memory is contained within the DtT ASR engine.
19. The audio processing method of claim 1 , wherein the receiving audio data step involves obtaining data directly froma voice-over Internet protocol (VOIP) telephony network using anactive recording protocol.
20. The audio processing method of claim 1 , wherein the audio receiving step utilizes a port mirroring switch to obtain data directly from a VoIP telephony network by packet sniffing.