Method and System for Providing Shift-Invariant Features for Improved Audio Analytics on Radio-Frequency (RF) Channels
Publication Date: 2014-Apr-07
The IP.com Prior Art Database
A method and system is disclosed for providing shift-invariant features for improved audio analytics on radio-frequency (RF) channels.
Page 01 of 1
Method and System for Providing Shift - -Frequency
Disclosed is a method and system for providing shift-invariant features for improved audio analytics on radio-frequency (RF) channels. The method and system follows segment based approach that segments audio signals to speech and non -speech regions by applying one or more shape detection techniques on corresponding spectrogram. A spectrogram displays the different frequencies present in audio signals .
Application of shape detection techniques on the spectrogram results in interpreting acoustic patterns of audio signals at different frequencies . The acoustic patterns are used to simultaneously determine the boundaries between different speech and non-speech segments and the corresponding labels.
In accordance with the method and system, before segmenting, an audio preprocessing algorithm such as, for example, Speech Activity Detection (SAD) algorithm is used for pre-processing the audio signals in RF channels. The SAD algorithm thus pre-processes the audio signals from RF channels and improves relative quality and performance of the speech/non-speech detection in terms of total error on four RF channels used in Robust Automatic Transcription of Speech (RATS) program.
In a scenario, after segmenting the audio signals to short segments based on detection of change points, the segments are represented using shift-invariant features. Thereafter, by employing a representation that is shift-in...