Audio signal coding with a signal adaptive filterbank

• US 5,732,189 A
• Filed: 12/22/1995
• Issued: 03/24/1998
• Est. Priority Date: 12/22/1995
• Status: Expired due to Term

• Current Assignee: Lucent Technologies, Inc. (Nokia Corporation)
• Original Assignee: Lucent Technologies, Inc. (Nokia Corporation)
• Inventors: Princen, John Peter, Johnston, James David
• Primary Examiner(s): MacDonald, Allen R.
• Assistant Examiner(s): Dorvil, Richemond
• Application Number:
  US08/577,350
• Time in Patent Office: 823 Days
• Field of Search: 395/2.39, 395/2.38, 395/2.36, 395/2.35, 395/2.33, 395/2.91, 395/2.92, 395/2.12, 395/2.2, 395/2.09
• US Class Current: 704/230
• CPC Class Codes: H04B 1/665 using psychoacoustic properties of the ear, e.g. masking effect

Abstract

An audio coding technique is presented, which utilizes a time varying non-uniform filterbank in conjunction with a Perceptual Audio Coder. The non-uniform filterbank is formed from a plurality of uniform filter bank sections, and a transition filter. Each uniform filterbank section covers a portion of a predetermined frequency axis, and the transition filter is utilized to cover the remainder of the predetermined frequency axis. Aliasing terms introduced at the band edges of the transition filter are cancelled to obtain an overall transfer function for the non-uniform filterbank, which has linear phase and is approximately flat. Use of the non-uniform filterbank permits an increase in temporal resolution, which results in an improved coding technique over a wide range of frequencies.

Claims

1. A method of digitally encoding a digital/audio input signal to generate a compressed digital bit stream comprising the steps of:

calculating a plurality of threshold noise levels in response to time frequency characteristics of the digital audio input signal,

selectively filtering said digital/audio input signal in a time varying non-uniform filterbank to generate a filtered signal, said time varying form filter bank having a plurality of uniformly spaced filterbank sections covering a portion of a predetermined frequency axis and a transition filter being selected to cover a remaining portion of said predetermined frequency axis, said selective filtering step being responsive to said calculated threshold noise levels,

quantizing said filtered signal to generate a quantized signal, said quantizing step being responsive to said threshold noise levels, and

encoding said quantized signal to generate said compressed digital bit stream.

2. A method in accordance with claim 1, wherein there is further included the steps of:

determining a quadratic error measure for said transition filter, and

optimizing said quadratic error measure so that said error measure is minimized.

3. A method in accordance with claims 2, wherein there is further included the step of canceling aliasing terms introduced at the band edges of said transition filter, whereby an overall transfer function for said non-uniform time varying filterbank is obtained, which has linear phase and is approximately flat.

4. A method in accordance with claim 3, wherein said transition filter is a modulated filter.

5. A method of digitally encoding a digital/audio input signal to generate a compressed digital bit stream comprising the steps of:

calculating a plurality of threshold noise levels in response to time-frequency characteristics of the digital/audio input signal,

dividing said digital/audio input signal into a plurality of frequency sub-bands by inputting said digital/audio signal into a time varying non-uniform filterbank comprised of a plurality of uniformly spaced filterbank sections with a selectively added transition filter,

generating one or more digital samples for each of said plurality of frequency subbands, the number of digital samples generated being responsive to selected ones of the plurality of frequency sub-bands,

quantizing said digital samples to generate a quantized digital signal, said dividing, generating and quantizing steps being responsive to said threshold noise levels, and

encoding said quantized digital signal to generate said compressed digital bit stream.

6. A method of digital encoding in accordance with claim 5, wherein the number of digital samples generated is greater when the frequency of the selected frequency sub-band is higher.