Abstract A new signal quantization scheme is proposed which reduces fluctuation of the output signal by a signal quantizer (10) providing a quantized output signal and multiplying (18) said input signal X by a factor (1-w 1 )and finding a difference signal Δ in (11) between both the input signal and the previous input signal and multiplying (16) that by a weighting factor w 2 from a control (13). The previous quantizer output signal Q prev is summed (17) with the weighted difference signal Δ in w 2 and the sum is weighted by a weighting factor w 1 at a multiplier (19) to yield w 1 (Q prev +w 2 Δ in ). This signal is then summed at an adder (21) and applied to the quantizer (10) so that the quantizer is forced to match the fluctuation in the input signal as well as the signal itself. When applied to speech encoding algorithms such as the North American digital cellular telephone standard VSELP, the new quantizer results in more natural sounding background noise.
In a VSELP speech encoder that is responsive to digital input speech signals s(n) to provide a compressed digital output signal including quantized frame energy signal R and quantized reflection coefficient signal r i , an improved quantizer including: covariance means for generating a covariance matrix signal including means for calculating a signal covariance matrix according to ##EQU9## frame energy generator means coupled to said covariance means for generating a frame energy signal R including means for calculating the frame energy according to ##EQU10## where n is length of summation; means for calculating weights w 1 +w 2 for reduced fluctuation by calculating T=1.25 (T'-0.2) ##EQU11## W.sub.1 =0.75 (1-T)(1-S) W.sub.2 =T first quantizing means responsive to said frame energy signal for quantizing said frame energy signal R with a first reduced fluctuation quantizer (B) to provide quantized frame energy signal R; spectral Smoothing means for spectral smoothing said signal covariance matrix signal to provide a spectral smoothed signal; means coupled to said spectral smoothing means for generating a reflection coefficient signal r i including means for calculating reflection coefficients for i=1 to 10; and means responsive to said reflection coefficient signal r i for quantizing signal r i using second reduced fluctuation quantizer (A) to provide quantized reflection coefficient signal r i .
Claim 2
Dependent Claim
213 words
In a VSELP speech encoder that is responsive to digital input speech signals s(n) to provide a compressed digital output signal including quantized frame energy signal R and quantized reflection coefficient signal r i , an improved quantizer including: covariance means for generating a covariance matrix signal including means for calculating a signal covariance matrix according to ##EQU9## frame energy generator means coupled to said covariance means for generating a frame energy signal R including means for calculating the frame energy according to ##EQU10## where n is length of summation; means for calculating weights w 1 +w 2 for reduced fluctuation by calculating T=1.25 (T'-0.2) ##EQU11## W.sub.1 =0.75 (1-T)(1-S) W.sub.2 =T first quantizing means responsive to said frame energy signal for quantizing said frame energy signal R with a first reduced fluctuation quantizer (B) to provide quantized frame energy signal R; spectral Smoothing means for spectral smoothing said signal covariance matrix signal to provide a spectral smoothed signal; means coupled to said spectral smoothing means for generating a reflection coefficient signal r i including means for calculating reflection coefficients for i=1 to 10; and means responsive to said reflection coefficient signal r i for quantizing signal r i using second reduced fluctuation quantizer (A) to provide quantized reflection coefficient signal r i .