JPH0632023B2 - Voice analyzer - Google Patents

Description

Description: TECHNICAL FIELD The present invention relates to a configuration of a voice analysis device, and more particularly to a voice analysis device of a type in which a voice spectrum component is extracted by a band filter group in a voice recognition device.

[Conventional technology]

2. Description of the Related Art Conventionally, as a voice analysis device used in a voice recognition device or the like, a method of analyzing a time change of a voice spectrum component by a band filter group as shown in FIG. 5 has been widely used. That is, the input voice signal is fed to the variable gain amplifier 1
The signal is amplified to an appropriate amplitude by, and then the high frequency component is emphasized by the equalizer 2 having a frequency characteristic of 6 dB / Oct. This is because the frequency spectrum of the audio signal is not constant, and the higher the frequency component is, the more the frequency spectrum is attenuated, and the slope is -6 dB / Oct. Next, an input signal is decomposed into N channel frequency components by a band filter group 3 composed of N-order time division multiplex switched capacitors corresponding to the number of divisions N of the frequency band to be analyzed, rectified by a rectifier 4, and A low-pass filter group 5 placed for each component converts it into a direct current, and an analog / digital converter 6 converts it into a digital signal. The bandpass filter group 3 can be configured as shown in FIG. 6, for example. Each switch is turned on and off as shown in FIG. That is, switches φ _o , φ _e and i which are turned on and off complementarily with a fixed clock.
Th N switches phi ₁ as switch phi _i when selecting bandpass filter is turned on _......, φ _i, ...... φ _N and, between the inputs of the operational amplifier A1 and the input terminal by the switches, or operational amplifier A1 , And the input of the operational amplifier A2, or between the input and output of the operational amplifier A1, or between the input and output of the operational amplifier A2, or a capacitor group (C _a1 , C _a2 , ..., C) that is grounded. _aN ,
C _b1 , C _b2 , ..., C _bN , C _c1 , C _c2 , ...
, C _cN , C ₁ , C ₂ , C ₃ , C ₄ ), the N-th order time division multiplex band filter is configured by the switched capacitor filter. The low-pass filter group 5 is also composed of a time division multiplex switched capacitor filter. As a result, the voice analysis device shown in FIG. 4 is realized by a monolithic IC.

[Problems to be solved by the invention]

In the above-described conventional voice analysis device, the input DC offset generated inside the filter due to the DC offset generated in the operational amplifier forming the bandpass filter group is multiplied by the coefficient determined by the filter characteristic and appears as the output DC offset. Since the band pass filter groups have different frequency characteristics, the value of the coefficient applied to the input DC offset is different, and the output DC offset is different for each band filter.
Therefore, the DC offset generated in each band filter is superimposed on the spectral component of the input voice and output to the output end of the voice analysis device. When the band-pass filter group is composed of N band-pass filters, the DC offset can be eliminated by connecting a capacitor between the output of each band-pass filter and the input of the rectifier, but the same number of capacitors as the band-pass filter group can be eliminated. Therefore, there is a disadvantage that the chip area becomes extremely large when the monolithic IC is formed. In particular, when implementing the bandpass filter group by a time division multiplexing method using a switched capacitor,
Since the magnitude of the DC offset differs depending on each filter characteristic, it is impossible to cut off the DC offset by the capacitor.

In a voice recognition device characterized by analyzing a voice spectrum by a bandpass filter group, a voice spectrum pattern output from a voice analysis device is compared with a plurality of reference voice spectrum patterns, and the most similar pattern. It is determined that is equal to the input voice pattern.
Therefore, when the voice spectrum output from the voice analysis device includes a DC offset, the original voice spectrum pattern includes an error, which results in a reduction in the recognition rate.

[Means for solving problems]

A speech analysis apparatus of the present invention comprises a variable gain amplifier for amplifying a speech input signal to a desired amplitude, an equalizer for correcting the frequency characteristic of the output signal of the variable gain amplifier, and N output signals of the equalizer. A band-division filter group of a time-division multiplexing system that divides the frequency band, a rectifier that rectifies the output signal of the band-filter group, a low-pass filter group of the time-division multiplexing system that converts the output signal of the rectifier into a DC signal, Analog for converting the output signal of the low-pass filter group into a digital signal
In a voice analysis device constituted by a digital converter, a time-division multiplexing high-pass filter group for cutting off a DC signal is connected between the band-pass filter group and the rectifier.

〔Example〕

Next, the present invention will be described with reference to the drawings. FIG. 1 is a block diagram showing an embodiment of the present invention. The audio signal is input to the variable gain amplifier 1 and adjusted so that the maximum amplitude is within the dynamic range. Next, a bandpass filter group 3 having a gain of +6 dB / Oct by the equalizer 2 and composed of a time division multiplex switched capacitor filter
Are decomposed into N frequency components. The output of the band-pass filter group 3 is input to the high-pass filter group 7, and the band-pass filter group 3
The DC offset generated in each band is cut off.
The output of the high-pass filter group 7 is rectified by the rectifier 4, converted into a DC signal by the low-pass filter group 5 composed of time division multiplex switched capacitor filters, and then converted into a digital signal by the analog / digital converter 6. It

The high pass filter group 7 is realized by a time division multiplex switched capacitor filter, and can be configured as shown in FIG. 2, for example. Each switch is turned on and off as shown in FIG. That is, the switches φ _o and φ _e that are turned on and off complementarily at a fixed clock and the N switches φ ₁ , such that the switch φ _i is turned on when the i-th high-pass filter is selected.
_{......, φ i, ......, φ} N and between the inputs of these operations and by connexion input terminal to the switch amplifier A, or are connected between the input and output of the operational amplifier A or capacitor group C is grounded,
_a1 , C _a2 , ..., C _aN , C _b1 , C _b2 , ...,
A switched capacitor filter having C _bN , C ₁ ) constitutes an Nth-order time-division multiple high-pass filter. Capacitor C is used to equalize the frequency characteristics of each high-pass filter.
_{a1, C a2, ......, C} aN each equal, also the capacitor _{C b1, C b2, ......,} C bN also each equal. Letting T be the period in which each of the switches φ ₁ , φ ₂ , ..., φ _N is turned on and off, the transfer function H (z) between the input voltage V _1N and the output voltage V _OUT is Becomes Here, i = 1, 2, ..., N, z = exp (j
ωT) and ω is the angular frequency of the input signal. If the loss in the pass band of each high-pass filter is 0 dB, the cutoff frequency f _c is Becomes The cut-off frequency f _c of the high-pass filter group 7 needs to be smaller than the minimum value of the center frequency of the band-pass filter group 3 in order to prevent loss of the output signal of the band-pass filter group 3. Since the cutoff frequency f _c is inversely proportional to the ratio of the integral capacitance C _ai and the Samplinge capacitance C ₁ , the cutoff frequency f _c is
_{When c} is small, the integration capacity C _ai and the sampling capacity C
The ratio of ₁ becomes large. Since the size of the sampling capacitor C ₁ is determined by the ratio accuracy of the capacitors, the value of the integral capacitor C _ai becomes large. Since the total number of the integrated capacitors C _ai is 2N, the area occupied by the capacitors is large, and the chip area is increased. FIG. 3 is a high-pass filter group that does not include the above-mentioned drawbacks.
That is, the switch φ in the high-pass filter group of FIG.
Resistors r ₁ and r ₂ for dividing the output voltage V _OUT of the operational amplifier A are connected between ₀ and the output terminal of the operational amplifier A. At this time, the transfer function H (z) between the input voltage V _IN and the output voltage V _OUT is Becomes The cutoff frequency f _c is Becomes Therefore, the cutoff frequency f _c is equal to the resistance r ₁ + r ₂
And inversely proportional to the ratio of the resistance r _1, by increasing the resistance r _{1 +} r ₂ the ratio of the resistance r _1, a small integrated capacitors, it is possible to construct a suitable high-pass filter group monolithically IC.

〔The invention's effect〕

As described above, according to the present invention, in a voice analysis device using a bandpass filter group, a highpass filter group that blocks a DC offset is connected between the output of the bandpass filter group and a rectifier, and each filter group is time-division multiplexed. By configuring with a switched capacitor filter, there is an effect that a direct current offset having a different magnitude generated in each band filter group is cut off, and a voice recognition device with higher accuracy and suitable for a monolithic IC is realized.

[Brief description of drawings]

1 is a block diagram showing an embodiment of the present invention, FIG. 2 is a first circuit diagram showing an example of the high-pass filter group 7 of FIG. 1,
FIG. 3 is a second circuit diagram showing an example of the high-pass filter group 7 of FIG. 1, FIG. 4 is a time chart showing the operation of the switches of FIGS. 2, 3, and 6, and FIG. Is a block diagram showing a conventional voice analysis device, and FIG. 6 is a circuit diagram showing an example of the bandpass filter group of FIG. 1 ... Variable gain amplifier, 2 ... Equalizer, 3 ... Bandpass filter group, 4 ... Rectifier, 5 ... Low-pass filter group, 6 ...
Analog-digital converter, 7 ... Bandpass filter group (φ ₁ , φ ₂ , ..., φ _N ) ... Time division multiplexing switch φ _o , φ _e ...... Switched capacitor switch (C _a1 , C _a2 , ......, C _aN , C _b1 , C _b2 , ...
…, C _bN , C _c1 , C _c2 ,…, C _cN , C ₁ , C
₂ , C ₃ , C ₄ ) ... Capacitors A, A 1, A 2 ... Operational amplifiers r ₁ , r ₂ ... Resistors

Claims (1)

[Claims] 1. A variable gain amplifier for amplifying a voice input signal to a desired amplitude, an equalizer for correcting frequency characteristics of an output signal of the variable gain amplifier, and an N channel frequency for the output signal of the equalizer. A band-division filter group of a time-division multiplexing system for dividing into bands, a rectifier for rectifying an output signal of the band-filter group, a low-pass filter group of a time-division multiplexing system for converting an output signal of the rectifier into a DC signal, and Analog that converts the output signal of the low-pass filter group into a digital signal
A voice analysis device comprising a digital converter, characterized in that a time division multiplex system high-pass filter group for blocking a DC signal is connected between the band-pass filter group and the rectifier.