JPH03212698A - signal processing device - Google Patents

Abstract

PURPOSE:To improve distinctness by providing respective means for estimating a pitch frequency, generating a window in accordance therewith to predict noise, removing the noise according thereto and stressing the pitch by using a window in the noise removed output. CONSTITUTION:An input signal is processed in an FFT processing section 1 to an input signal (SF)'' converted to a frequency region. This SF is subjected to the detection of a cepstrum peak (KP) by a KP detecting section 2 and further a pitch frequency (PF) is determined by a PF estimating section 3. A window (WD) generating section 4 generates WD for voice stressing as the data of a frequency region in accordance with this PF and sends this data to a PF stressing section 7. The noise prediction of the SF is executed by a noise predicting section 5. This noise component is determined in a frequency region and is sent to a canceling section 6. The noise component is accurately removed together with the frequency component in this section and is sent to the stressing section 7. The signal subjected to noise removal in this section is subjected to voice stressing according to the WD from the generating section 4 and the signal of the stressing section 7 is converted to the signal of the time region in an IFFT processing section 8 and is outputted. The distinctness is improved in this way.

Description

DETAILED DESCRIPTION OF THE INVENTION Field of the Invention The present invention relates to a signal processing device related to noise removal processing of audio signals.

2. Description of the Related Art In recent years, noise removal methods using signal processing techniques have been widely studied.

FIG. 2 is a block diagram of a conventional signal processing device.

As shown in the figure, the fill control unit 11 has BPF group 1 upper group 2
The fill coefficient is controlled according to the noise of the input signal, and it is a BPF group 1 group and 2 pass filter groups, which divides the input signal into appropriate bands and controls the fill coefficient according to the control signal of the fill control section 11. , is configured to determine its passband characteristics.

The operation of the conventional signal processing device configured as described above will be described below.

The input signal in which noise is superimposed on the voice is supplied to the fill control unit 11, and the fill control unit 11 calculates the miscellaneous writing component from the human input signal according to the upper 2 bands of BPF group 1, and
BP fill coefficient that does not allow the two tone components to pass through.
Group F and Group 1 are given 2 doses.

BPF group 1 group 2 input signals are divided into appropriate bands, and the input signals are passed through according to the filter coefficients inputted from the fill control section 11 for each band and supplied to the adder 13. The adder 13 obtains an output obtained by mixing signals divided into two appropriate bands on the first BPF group. Therefore, since the output signal has a lowered pass level in the noise-containing band of BPF group 1 and upper 2, a signal with attenuated noise components can be obtained.

Problems to be Solved by the Invention However, in such conventional signal processing devices, S/
Although there is an improvement in the N ratio, there is a problem in that the clarity is not good. The present invention has been made to solve the above problems, and it is an object of the present invention to provide a signal processing device that can remove noise with good clarity.

Means for Solving the Problems In order to achieve the above objects, the present invention provides FFT processing means for converting an input signal into a frequency domain and outputting a frequency domain signal, and cepstrum analysis of the frequency domain signal to generate a cepstrum signal. cepstrum peak inspection means for determining a peak; pitch frequency estimating means for determining a pitch frequency from the cepstrum peak; window generating means for generating a frequency domain window signal using pitch information of the pitch frequency estimating means; and the FFT processing. noise prediction means for predicting noise from the frequency domain signal supplied by the means; canceling means for removing noise from the frequency domain signal supplied from the FFT processing means; and canceling means using the window signal of the window generation means. The present invention is configured to include pitch frequency emphasizing means for controlling the signal supplied from the pitch frequency emphasizing means, and IFFT processing means for converting the signal supplied from the pitch frequency emphasizing means into the time domain.

Operation With the above-described configuration, an input signal containing noise is converted into a frequency domain signal by the FFT processing means.

From the signal in the frequency domain, a cepstrum peak is determined by a cepstrum peak detection means, and a pitch frequency is further obtained by a pitch frequency estimation means. Based on the pitch frequency, the window generating means generates a window as frequency domain data for pitch emphasis.

Further, the noise of the input signal converted into a frequency domain signal by the FFT processing means is determined in the frequency domain by the noise prediction means. Then, the noise-containing input signal converted into a frequency domain signal by the FFT processing means is subjected to noise removal in the frequency domain by the canceling means based on the noise obtained by the noise prediction means.

Furthermore, the pitch frequency of the noise-removed input signal is emphasized by the pitch frequency emphasizing means in the frequency domain based on the window, by the canceling means;
The frequency domain signal is converted into a time domain signal by the IFFT processing means and output.

EXAMPLE An example of the present invention will be described below with reference to the drawings.

FIG. 1 shows a block diagram of a signal processing device in an embodiment of the present invention.

As shown in the figure, an FFT processing section 1 converts an input signal into a frequency domain signal, and a cepstrum peak detection section 2. Noise prediction unit 6. The signal is supplied to the cancellation section 6.

The cepstrum peak detection section 2 detects a cepstrum peak from the frequency domain signal obtained from the FFT processing section 1 and supplies it to the pitch frequency estimation section 3 .

The pitch frequency estimating section 3 calculates a pitch frequency from the cepstrum peak and supplies it to the window generating section 4, and the window generating section 4 generates a window according to the pitch frequency and supplies it to the pitch frequency emphasizing section 7. It has become. The noise prediction unit 5 performs noise prediction on the signal supplied from the FFT processing unit 1 and supplies the noise prediction signal to the canceling unit 6, and the canceling unit 6 performs noise prediction on the signal supplied from the FFT processing unit 1. The supplied signal is processed and supplied to the pitch frequency emphasizing section 7. The pitch frequency emphasizing section 7 performs pitch frequency emphasizing processing on the signals supplied from the window generating section 4 and the canceling section 6, supplies the result to the IFFTFF, converts the signal into a time domain signal, and outputs the signal to the IFFTFF. The structure is as follows.

To explain the operation of the above configuration, first, an input signal to the apparatus is converted into a frequency domain signal by the FFT processing section 1. The input signal transformed into the frequency domain is
The cepstrum peak detection unit 2 detects the cepstrum peak, and the pitch frequency estimation unit 3 further detects the cepstrum peak.
The pitch frequency can be found. Based on the determined pitch frequency, the window generating section 4 generates a suitable window for voice enhancement as frequency domain data and supplies it to the pitch frequency emphasizing section 7. Further, the noise prediction unit 5 predicts the noise of the input signal converted into the frequency domain, and calculates the noise component in the frequency domain and supplies it to the cancellation unit 6. And cancel part 6
Now, from the input signal supplied from the FFT processing unit 1 and converted into a frequency domain signal, the noise components in the frequency domain obtained by the noise prediction unit 5 are accurately removed for each frequency component, and the pitch is calculated. Supplies to frequency emphasis section.

Then, in the pitch frequency emphasizing section 7, the noise-removed frequency signal obtained from the canceling section 6 is controlled in accordance with the window for performing speech enhancement obtained from the window generating section 4, and is subjected to I FFT processing. Part 8
supply to. Then, the IFFT processing unit 8 converts the signal from the pitch frequency emphasizing unit 7 into a time domain signal and outputs the signal.

As described above, according to the signal processing device of the embodiment of the present invention, it is possible to remove noise from a signal in which noise is superimposed on speech, and also to emphasize the speech component by providing the pitch frequency emphasizing section. , it is possible to obtain an audio signal with excellent clarity.

It goes without saying that it is desirable that the window generated by the window generator 4 in the above embodiment be a window that represents the harmonic structure of the audio.
A combination of a comb filter and a low-pass filter may also be used. Furthermore, it goes without saying that the pitch frequency emphasizing section 7 can be easily implemented using a multiplication circuit.

Effects of the Invention As is clear from the above embodiments, the present invention provides pitch frequency estimation and flattening for estimating pitch frequency in a device that converts a signal into the frequency domain and performs noise removal processing, and generates a window according to the pitch frequency. a noise prediction means, a cancellation means for removing noise according to the output of the noise prediction means, and a pitch frequency emphasis means for emphasizing the pitch by using the window of the window generation means for the cancellation output of the cancellation means. By providing this, it is possible to remove noise from a signal in which noise is superimposed on speech and further emphasize the speech component, thereby making it possible to obtain a speech signal with high clarity.

[Brief explanation of drawings]

FIG. 1 is a block diagram of a signal processing device according to an embodiment of the present invention, and FIG. 2 is a block diagram of a conventional signal processing device. 1...FFT processing unit, 2...Cepstrum peak detection unit, 3...Pitch frequency estimation unit, 4...Window generation unit, 6... ...Noise prediction unit, 6... Self-cancellation unit, 7...
Pitch frequency emphasis unit, 8...IFFT processing unit.

Claims (1)

[Scope of Claims] FFT processing means for performing FFT processing on a signal containing noise; cepstrum peak detection means for performing cepstrum analysis of the FFT output of the FFT processing means and detecting its peak; and a peak detection output of the cepstrum peak detection means. a pitch frequency estimating means for estimating the pitch frequency of the FFT processing means; a window generating means for generating a window output based on the pitch frequency estimation output of the pitch frequency estimating means;
a noise prediction means for predicting noise from the FT output; and the FF
means for canceling the noise prediction output of the noise prediction means from the FFT processing output of the T processing means; pitch frequency emphasis means for multiplying the canceled output of the cancellation means by the window output of the window generation means; and enhancement of the pitch frequency emphasis means. A signal processing device equipped with an IFFT processing means for performing IFFT processing on an output.