Obviousness

Under 35 U.S.C. § 103, a patent claim is considered obvious if the differences between the claimed invention and the prior art are such that the subject matter as a whole would have been obvious at the time the invention was made to a person having ordinary skill in the art (POSITA) to combine existing elements from the prior art. The motivation to combine references can come from the references themselves, from common sense, or from the knowledge of a POSITA.

The core invention of US Patent 6,807,524, as outlined in independent claims 1 and 8, involves a perceptual weighting device and method that utilizes a signal pre-emphasis filter, a synthesis filter calculator responsive to the pre-emphasized signal, and a perceptual weighting filter with a "fixed denominator" to decouple formant weighting from spectral tilt for wideband speech signals. Further claims (e.g., claims 4 and 5) specify the pre-emphasis filter as P(z)=1-μz⁻¹ and the perceptual weighting filter as W(z)=A(z/γ₁)/(1-γ₂z⁻¹), with the critical condition that γ₂ is set equal to μ.

Obviousness Analysis

A person having ordinary skill in the art (POSITA) in speech coding at the time of the invention (priority date 1998-10-27) would likely have been motivated to combine the following prior art references and general knowledge to arrive at the claimed invention:

Prior Art References and Teachings:

1. Atal and Schroeder, "Predictive coding of speech and subjective error criteria" (1979): This foundational paper, cited within US6807524, describes the traditional perceptual weighting filter in analysis-by-synthesis (AbS) coders, which has the transfer function W(z) = A(z/γ₁)/A(z/γ₂). It also explains that the quantization error is weighted by W⁻¹(z) to exploit the masking property of the human ear [cite: US6807524 Description, Section "Perceptual Weighting", paragraph 2].
2. Admitted Prior Art Problem and Suggestion (US6807524 Background): The patent itself explicitly states that the traditional perceptual weighting filter W(z) = A(z/γ₁)/A(z/γ₂) "is not suitable for efficient perceptual weighting when it was applied to wideband signals" because it has "inherent limitations in modelling the formant structure and the required spectral tilt concurrently." Crucially, the patent admits that "It was suggested to add a tilt filter into W(z) in order to control the tilt and formant weighting separately" [cite: US6807524 Description, Section "Brief description of the prior art", paragraph 6]. This provides both the known problem and a general prior art solution strategy.
3. General Knowledge of Pre-emphasis and De-emphasis Filters in Speech Coding: Pre-emphasis filters of the form P(z) = 1 - μz⁻¹ are widely known and commonly employed at the input of speech encoders to enhance high-frequency content, reduce the dynamic range of the signal, and improve the stability and accuracy of linear prediction (LP) analysis. The corresponding de-emphasis filter D(z) = 1/(1 - μz⁻¹) is used at the decoder output. This is described as a conventional technique in the background of US6807524 [cite: US6807524 Description, module 103 description].
4. General Knowledge of Performing LP Analysis on Pre-emphasized Signals: It is a known practice in speech coding to perform LP analysis on a pre-emphasized version of the speech signal to derive the LP filter coefficients A(z). This is also explicitly described in US6807524 [cite: US6807524 Description, module 104 description].

Motivation for Combination and Obviousness:

A POSITA, aware of the limitations of the conventional perceptual weighting filter W(z) = A(z/γ₁)/A(z/γ₂) for wideband signals—specifically its difficulty in concurrently modeling formant structure and spectral tilt—would be motivated by the admitted prior art suggestion to "add a tilt filter into W(z) in order to control the tilt and formant weighting separately" [cite: US6807524 Description, Section "Brief description of the prior art", paragraph 6].

Given that a first-order pre-emphasis filter P(z) = 1 - μz⁻¹ is a common and simple "tilt filter" used to shape the spectral slope of a speech signal, a POSITA would find it obvious to modify the denominator of the conventional perceptual weighting filter A(z/γ₂) to a simpler, fixed-pole, first-order filter (1-γ₂z⁻¹) to directly implement the suggested "tilt filter." This change effectively decouples the adaptive formant shaping provided by the numerator A(z/γ₁) from a fixed spectral tilt control in the denominator. Such a simplification would be a logical design choice for a POSITA attempting to address the known problem of concurrent modeling.

Furthermore, in a complete speech coding system that already incorporates a pre-emphasis filter P(z)=1-μz⁻¹ at the encoder input and a de-emphasis filter D(z)=1/(1-μz⁻¹) at the decoder output, a POSITA would consider the overall error shaping. The total error shaping would be determined by the inverse of the weighting filter multiplied by the inverse of the pre-emphasis filter (W⁻¹(z)P⁻¹(z)). To simplify this overall error shaping and achieve a desired and perceptually optimized error spectrum, such as one directly related to the inverse of the analysis filter (1/A(z/γ₁)), a POSITA would have a clear motivation to set the pole of the fixed denominator (γ₂) in the modified weighting filter W(z) = A(z/γ₁)/(1-γ₂z⁻¹) to be equal to the pre-emphasis factor (μ). This specific setting (γ₂ = μ) causes the term (1-γ₂z⁻¹) in the denominator of W(z) to cancel with the P⁻¹(z) term from the de-emphasis filter in the overall error shaping, resulting in W⁻¹(z)P⁻¹(z) = 1/A(z/γ₁). The patent itself highlights that this particular configuration "is very efficient for encoding wideband signals, in addition to the advantages of ease of fixed-point algorithmic implementation" [cite: US6807524 Description, Section "Perceptual Weighting", paragraph 5], thereby providing strong evidence of its desirability and the motivation for a POSITA to achieve it.

Therefore, the combination of:

• The known perceptual weighting filter (Atal and Schroeder).
• The admitted prior art problem with this filter in wideband applications and the suggestion to add a tilt filter (US6807524 Background).
• The common use of pre-emphasis (P(z)=1-μz⁻¹) at the encoder input and performing LP analysis on the pre-emphasized signal (General knowledge/US6807524 Background).
• And the understanding of the interaction between pre-emphasis and the weighting filter in error shaping, leading to the motivation to set γ₂ = μ to simplify the effective error weighting to 1/A(z/γ₁) (US6807524 Description).

Would render the claimed device, method, and system of US Patent 6,807,524 obvious to a POSITA. The specific value of μ=0.7 (claim 3) is a typical design parameter for pre-emphasis filters, not an inventive limitation, once the structure is obvious.