Method and device for spectral expansion for an audio signal

• US 10,043,534 B2
• Filed: 12/22/2014
• Issued: 08/07/2018
• Est. Priority Date: 12/23/2013
• Status: Active Grant

• Current Assignee: Staton Techiya, LLC
• Original Assignee: Staton Techiya, LLC
• Inventors: Usher, John, Ellis, Dan
• Primary Examiner(s): Yang, Qian
• Application Number:
  US14/578,700
• Publication Number:
  US 20150179178A1
• Time in Patent Office: 1,324 Days
• Field of Search: None
• CPC Class Codes: G10L 21/038 using band spreading techniques

Abstract

A method and device for automatically increasing the spectral bandwidth of an audio signal including generating a “mapping” (or “prediction”) matrix based on the analysis of a reference wideband signal and a reference narrowband signal, the mapping matrix being a transformation matrix to predict high frequency energy from a low frequency energy envelope, generating an energy envelope analysis of an input narrowband audio signal, generating a resynthesized noise signal by processing a random noise signal with the mapping matrix and the envelope analysis, high-pass filtering the resynthesized noise signal, and summing the high-pass filtered resynthesized noise signal with the original an input narrowband audio signal. Other embodiments are disclosed.

Claims

1. A method for automatically expanding the spectral bandwidth of an audio signal comprising:

generating a mapping matrix based on an analysis of a reference wideband signal and a reference narrowband signal, the mapping matrix being a transformation matrix to predict high frequency energy from a lower frequency energy envelope, where the mapping matrix is generated without using a linear predictive coefficient (LPC) method, wherein the mapping matrix is generated based, in part, on using a dB domain for performing a linear prediction;

generating an energy envelope analysis of an input narrowband audio signal;

generating a resynthesized noise signal by processing a random noise signal with the mapping matrix and the energy envelope analysis;

high-pass filtering the resynthesized noise signal;

generating an output audio signal by summing the high-pass filtered resynthesized noise signal with the input narrowband audio signal; and

generating an audible output from a speaker using the output audio signal.

2. The method of claim 1, wherein the reference wideband and reference narrowband signals are made from a simultaneous recording of a sentence made with an ambient microphone and an ear canal microphone located in an earphone.

3. The method of claim 1, where the input narrowband audio signal is taken from an ear-canal microphone located in an earphone.

4. The method of claim 1, where the input narrowband audio signal is taken from a received speech audio signal in a speech telecommunications system.

5. The method of claim 1, where the summed signal is directed to a speech telecommunications system.

6. The method of claim 1, where the summed signal is directed to a voice controlled device.

7. The method of claim 1, where the mapping matrix is generated from a least squares fit analysis of the reference wideband and reference narrowband signals.

8. The method of claim 1, where the mapping matrix is generated by a linear regression model, where the input reference wideband signals and reference narrowband signals are first converted to a frequency domain representation, and secondly each frequency band envelope is converted to a dB domain representation.

9. A non-transitory computer readable medium containing instructions for spectral enhancement, the execution of the instructions by a processor of a computer system causing the processor to perform operations comprising:

generating a mapping matrix based on an analysis of a reference wideband signal and a reference narrowband signal, the mapping matrix being a transformation matrix to predict high frequency energy from a lower frequency energy envelope, where the mapping matrix is generated without using a linear predictive coefficient (LPC) method, wherein the mapping matrix is generated based, in part, on using a dB domain for performing a linear prediction;

generating an energy envelope analysis of an input narrowband audio signal;

generating a resynthesized noise signal by processing a random noise signal with the mapping matrix and the energy envelope analysis;

high-pass filtering the resynthesized noise signal;

generating an output audio signal by summing the high-pass filtered resynthesized noise signal with the input narrowband audio signal; and

generating an audible output from a speaker and using the output audio signal.