Keyword Spotting Engine

With the arrival of digital communication, the quantity of audio files being created is growing ever more quickly and shows no signs of slowing.  With billions of hours of audio per month, valuable information exist in these audio and innovative ideas are being attempted and designed to utilize such data.  However, effectively retrieving information from such massive audio data has proven to be difficult. Traditional techniques have proven to be not feasible for such task.  For example, traditional speech-to-text engine, such as our own Hosted Speech Recognition and Dragon SDKs, would require hundreds and thousands of computers and would cost millions dollars per month to operate alone.

 

Introduce Nexiwave Keyword Spotting Engine.  The Nexiwave Keyword Spotting engine searches the occurrences of a list of keywords or key phrases in audio stream effectively and accurately.  With speed thousands times faster than realtime, it is capable of processing millions hours of audio per day. More importantly, only a single modern CPU core is required to run an instance of Nexiwave Keyword Spotting engine and less than 10mb memory for each audio stream. This will leave plenty of precious CPU resources for other tasks, such as your business logic or another instance of Nexiwave Keyword Spotting engine.

Applications:

  • Call Center Quality Control: a call center manager can enroll a list of keywords or phrases and instantly notified when any of such words are spoken in the calls.  Proper actions, such as direct intercept into the call, can then be taken.
  • Call Recorder Embedding: with the very low CPU and memory footprints, the Nexiwave Keyword Spotting engine can be easily embed in a call recorder.
  • Instant Call Alert: Phone calls may be monitored for list of keywords. Instant alerts can be easily raised when any of the keywords were spoken.  Archived calls can be easily re-searched for any previously not-enrolled keywords.
  • Archived Media Search: it is now possible to search the list of locations of certain keywords and jump directly to the point of interests.

Features:

  • Very high performance: process 1.2 millions hours per day on a single CPU core
  • Accurate keyword detection: average 70% detection rate
  • Detection delay: none, near instant
  • Low memory footprint: less than 10mb memory per audio stream
  • Embeddable to call recorders
  • Host on your own or embed into your recorder

Target Users:

  • Speech Analytic developers
  • Large volume audio search developers

Contact us today for Nexiwave Keyword Spotting technology.


Terms of Use    © Copyright 2017 Nexiwave
Email:
   legal  
  legal  cancel