System and method to correct for packet loss in automatic speech recognition systems utilizing linear interpolation

• US 10,157,620 B2
• Filed: 03/04/2015
• Issued: 12/18/2018
• Est. Priority Date: 03/04/2014
• Status: Active Grant

• Current Assignee: Genesys Telecommunications Laboratories Incorporated (Genesys Cloud Services Incorporated)
• Original Assignee: Interactive Intelligence Group Incorporated (Genesys Cloud Services Incorporated)
• Inventors: Cheluvaraja, Srinath, Iyer, Ananth Nagaraja, Ganapathiraju, Aravind, Wyss, Felix Immanuel
• Primary Examiner(s): WOZNIAK, JAMES S
• Application Number:
  US14/638,198
• Publication Number:
  US 20150255075A1
• Time in Patent Office: 1,385 Days
• Field of Search: 704201, 704222, 714747
• CPC Class Codes: G10L 15/02 Feature extraction for speech recognition; Selection of recognition unit
• CPC Class Codes: G10L 15/08 Speech classification or search
• CPC Class Codes: G10L 15/14 using statistical models, e.g. Hidden Markov Models [HMMs] G10L15/18 takes precedence
• CPC Class Codes: G10L 15/142 Hidden Markov Models [HMMs]
• CPC Class Codes: G10L 15/20 Speech recognition techniques specially adapted for robustness in adverse environments, e.g. in noise, of stress induced speech G10L21/02 takes precedence
• CPC Class Codes: G10L 19/005 Correction of errors induced by the transmission channel, if related to the coding algorithm
• CPC Class Codes: G10L 2015/025 Phonemes, fenemes or fenones being the recognition units
• CPC Class Codes: G10L 2019/0012 Smoothing of parameters of the decoder interpolation
• CPC Class Codes: G10L 25/18 the extracted parameters being spectral information of each sub-band
• CPC Class Codes: G10L 25/21 the extracted parameters being power information

Abstract

A system and method are presented for the correction of packet loss in audio in automatic speech recognition (ASR) systems. Packet loss correction, as presented herein, occurs at the recognition stage without modifying any of the acoustic models generated during training. The behavior of the ASR engine in the absence of packet loss is thus not altered. To accomplish this, the actual input signal may be rectified, the recognition scores may be normalized to account for signal errors, and a best-estimate method using information from previous frames and acoustic models may be used to replace the noisy signal.

Claims

1. A method to correct for packet loss in an audio signal in an automatic speech recognition system comprising reconstructing the signal using linear interpolation from preceding and succeeding frames of audio, wherein the linear interpolation comprises the steps of:

a. determining spectral vectors from a time interval of frames preceding and succeeding a packet loss event, the packet loss event comprising two to three frames;

b. maintaining a sliding buffer, comprising a number of frames, preceding and succeeding the packet loss event and a number of frames of the packet loss event;

c. generating the linear interpolation for the time interval of frames, wherein the two to three frames of the packet loss event undergo linear interpolation simultaneously using the frames preceding and succeeding the packet loss event in the sliding buffer;

d. using energy of the signal to tag frames within the packet loss event;

e. inserting values for the tagged frames within the packet loss event into the sliding buffer, the values being generated from the linear interpolation; and

f. inserting generated frames for the tagged packet loss event into the automatic speech recognition system.

2. The method of claim 1, wherein the sliding buffer comprises 10 frames.

3. The method of claim 1, wherein the packet loss event comprises digital silence.

4. The method of claim 1, wherein the packet loss event comprises three successive frames of digital silence.

5. The method of claim 1, wherein the time interval of frames are decomposed into overlapping frames of 20 ms with an overlap factor.

6. The method of claim 1 wherein the frames of the sliding buffer preceding the packet loss event comprise at least two frames,

wherein the frames of the sliding buffer succeeding the packet loss event comprise at least two frames, and

wherein the linear interpolation is generated from the at least two frames of the preceding the packet loss event and the at least two frames succeeding the packet loss event.