JPH01315797A - Pitch extractor - Google Patents

Abstract

PURPOSE:To perform stable and smooth pitch extraction in a wide pitch range by multiplying inputted acoustic signals by stability and inputting the multiplied results to a pitch extracting means as signals which are more emphasized in large amplitude and low frequency. CONSTITUTION:Upon receiving digital acoustic signals from an A/D converter 1, an EC value calculating section 4 calculates an EC value. The EC value is the total sum of the sample values of all sampling points between two continuous zero-crossing points in the signals. The EC value is inversely proportional to the frequency, but proportional to the amplitude. By utilizing such characteristics, therefore, the EC value is multiplied by the original digital signal outputs by means of a multiplier circuit 6 for calculating stability. As a result, no output is outputted at the time of no sound, but outputs which are more emphasized in large amplitude and low frequency are outputted from a multiplier circuit 5 when sounds exist. A self-correlating section 7 calculates self-correlating functions and outputs the calculated functions to a pitch discriminator at every sample. A pitch deciding section 8 estimates a pitch period from the output of the section 7. If the stability is high when a pitch abruptly changes, the pitch output is outputted to the outside.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a pitch extracting device for extracting the pitch, that is, the pitch period, pitch frequency, or pitch time of sounds such as musical instrument performance sounds or singing sounds.

[Prior art] Acoustic waveforms such as musical sounds and voices, except for noisy sounds such as unvoiced sounds, mostly have periodic repeating waveforms, and the change characteristics of the period, or pitch period, can be determined by acoustic analysis and synthesis. Or it becomes an important parameter in recognition, etc. For example, in an acoustic analysis and synthesis system, the pitch extraction result extracted by the analysis section has a large effect on the quality of the synthesized sound synthesized by the synthesis section.

Conventionally, various pitch extraction methods are known as methods for extracting the pitch period of an acoustic signal waveform, such as a method of calculating and extracting an autocorrelation coefficient for each frame having a time length of about the pitch period (for example, JP-A No. 63-23200, W, He
5s, “Pitch Determination
ofspeech jignal -, Sprlng
er-j/erlag, 4983, and Fujisaki et al.
``A new method for extracting the fundamental frequency based on the continuous and intermittent portions of speech waveforms'', Speech Study Group Materials, 5P86-95, etc.).

The pitch extraction method based on the autocorrelation coefficient is widely used because the autocorrelation coefficient can be calculated using a time-domain algorithm, and the influence of the phase between the waveform to be decomposed and the frame is relatively small. It is used.

Such pitch extraction methods are also a central theme in music recognition, and various commercial pitch extraction devices have already been devised (e.g., IVL, Pitch Rider).
5eries%FalrLight, VoiceTr
acker, Roland, VolceProces
sor, MIDI Guitar, and casto
Commercial pitch extraction devices such as MIDI Guitar, Inc., MIDI Guitar, etc.
This is achieved by converting the information into NoteloN/OFF information (Digital Interface), pitch bend information, etc., and connecting a MIDI sound source to the subsequent stage.

[Problems to be Solved by the Invention] By the way, in conventional pitch extraction devices, pitch overtone components, double pitch components, harmonic components other than pitch, etc. may cause erroneous extraction, which is a problem. . In order to avoid such erroneous extraction, it is known to limit the pitch search range (emphasizing smoothness) and to remove unnecessary frequency components in the previous stage.

However, conventional pitch extraction devices are limited to audio (speech).
Many of them assume a pitch range of about 80 to 300 Hz, and it is sufficient to perform a filter operation before pitch extraction, thereby removing unnecessary harmonic components and extracting a smooth pitch locus. On the other hand, in the case of music performance sounds, the pitch range is wide, about 40 to 1200 Hx, and the conventional extraction techniques described above have a problem in that high pitch parts cannot be extracted.

That is, in the case of music performance sounds, unlike in the case of normal voices, it is necessary to deal with unique latent elements such as sudden pitch changes and high pitch sounds.

Furthermore, there is a problem in that in a small amplitude section included in a signal wave, the touch excitation tends to become unstable, making pitch estimation unstable.

SUMMARY OF THE INVENTION An object of the present invention is to provide a pitch extraction device that can stably extract pitches of sound over a wide range in view of the problems of the conventional type described above.

[Means for Solving the Problems] In order to achieve the above object, the pitch extraction device according to the present invention calculates the stability such that the larger the amplitude and the lower the frequency, the higher the value. The stability is calculated by means 301, the product of this stability and the human acoustic signal is obtained by multiplication means 302, and this product signal is calculated by known pitch extraction means 30.
3 to extract the pitch.

[Operation] According to the above-described configuration, the input acoustic signal is transmitted to the multiplier 302.
is multiplied by the stability. Therefore, this multiplication means 3
The product signal output from 02 has a larger amplitude as the stability is higher, and a smaller amplitude as the stability is lower. Pitch extraction means 303
performs pitch extraction based on this product signal.

The stability shows a higher value as the input acoustic signal has a larger amplitude and a lower frequency. Therefore, the multiplier 302 suppresses the high frequency and small amplitude parts of the input acoustic signal, and emphasizes the large amplitude and low frequency. The signal will be input to pitch extraction means 303. Pitch extraction means 303 performs pitch extraction based on this signal.

[Example] Hereinafter, the present invention will be explained in detail using the drawings.

FIG. 2 is a block diagram showing the configuration of a pitch extraction device according to a first embodiment of the present invention. FIG. 3 is a block diagram showing an example of the circuit of the noise level determination section 2 in FIG.
The figure is a block diagram showing an example of a circuit of the post-processing section 9.

The operation of the apparatus of this embodiment will be explained with reference to FIGS. 2 to 4.

When audio signals (analog) such as voices and music are input, the A/D converter 1 first digitizes them. The digital audio signal sequence includes a noise level determination section 2 and a multiplication circuit 6.
, is output to the gate 3 and the EC value calculation section 4.

The noise level determination unit 2 inputs this human-powered acoustic signal and
The input acoustic signal is compared with the background noise level, and a signal indicating whether or not there is silence is outputted to the gate 3.

The operation of the noise level determining section 2 will be explained with reference to FIG. The noise level determination section 2 receives the power-on signal and holds the output level of the A/D converter 1 (FIG. 2) at that time in a hold circuit 21. This is the background noise level. Note that the background noise level may be measured over several seconds when the power is turned on.Also, the first measured result is used as the initial value of the background noise level, but from now on, this value will be used as the input signal. It may be changed adaptively depending on the situation.

The comparison circuit 22 compares the input acoustic signal with the hold circuit 21.
Compare the background noise level from Then, when the human acoustic signal is 1.4 times the background noise level or less (this value can be adjusted by the user), it is determined to be silent, and a signal indicating that there is silence is output during this silent section. do. Furthermore, the background noise level may be newly determined from the acoustic signal level value when it is determined that there is no sound and the past background noise level value.

Referring to FIG. 2, a signal indicating whether or not there is silence from noise level determining section 2 is input to gate circuit 3. As a result, gate 3 is turned off when there is no sound, and A/D converter 1
The digital audio signal output is not input to the multiplication circuit 5.

Next, the operation of EC value calculation 4 will be explained. EC value calculation 4 receives digital audio signal output from A/D converter 1 and calculates E
Calculate the C value. What is EC value?Execution Cy
It is an abbreviation for cle value, and refers to the sum of sample values of all sample points existing between two consecutive zero-crossing points in a signal.

FIG. 5(a) shows the continuous acoustic signal Sc sent to the A/D converter 1.
2 is a graph showing how a sample value SO, which is a digital audio signal, is obtained by sampling at a predetermined sampling interval. Among the sample values obtained in this way, those existing between two zero crossing points, for example, the sum of x1 to X1+4 in Fig. 5(b) is calculated as ECJ -Xl + The EC value is generally inversely proportional to the frequency and proportional to the amplitude, and the device of this embodiment utilizes this characteristic to improve the reliability of pitch extraction. .

Referring again to FIG. 2, the EC value calculated by the EC value calculation 4 is multiplied by the original digital signal output by the multiplication circuit 6. Thereby, stability is calculated.

Stability refers to the stability of the extraction state of the pitch extraction device, and is a function that can be said to be a measure of the reliability of the extracted results.

The EC value has a property of being inversely proportional to the frequency. Therefore, for signals with the same amplitude and different frequencies, the EC value will be larger for the low frequency signal. As described above, as the high frequency component of the signal wave increases, the possibility of incorrect pitch extraction increases, so the EC value can be an element of the stability function.

Furthermore, the EC value also has the property of being proportional to the input amplitude. Therefore, for signals with the same frequency but different amplitudes, the EC value will be larger for large amplitude waves. This property makes it possible to better reflect the situation where small amplitude signals are often accompanied by unstable pitch fluctuations. On the other hand, the EC value may locally decrease due to the influence of pitch overtone components, and some kind of correction is required. In this embodiment, the multiplication circuit 6 multiplies the original digital signal output by the EC value. This alleviates local fluctuations. Note that the value by which the EC value is multiplied may be an average amplitude value over a predetermined period of time of the digital audio signal.

Stability is calculated based on the EC value with the characteristics described above, so when a large amplitude, low frequency acoustic signal is input, it will inevitably show a large value, and conversely, if a small amplitude, high frequency sound signal is input, it will show a large value. In the case of frequency, it will show a small value.

The stability is output to the post-processing section 9 and also to the multiplication circuit 5. The multiplier circuit 5 multiplies the digital data string of the acoustic signal that is the output of the gate 3 by the stability calculated as described above, and as a result, the output is zero when there is no sound, and the output is large and low frequency is when there is sound. A more emphasized output is output from the multiplication circuit 5.

The autocorrelation unit 7 calculates and adds up the autocorrelation function of the human signal sequence for each sample, and outputs it to the subsequent pitch determiner for each frame period. FIG. 6 is a graph showing an example of the results of calculating the autocorrelation function.

Here, the autocorrelation function was calculated by the autocorrelation function calculation method using the formula Σ x , 2 plates=O. Although a method using semi-infinite exponential decay may be used, especially when the frame period is long, the autocorrelation function calculation method is more advantageous in terms of calculation cost.

The pitch determination section 8 estimates the pitch period from the output of the autocorrelation section 7 at the previous stage. Basically, the main processing contents are detection of the maximum peak position and quadratic interpolation for improving pitch accuracy. Here, ′ further gives the following constraint conditions (judgment conditions).

- The pitch search range is from ÷400 cents of the previous frame pitch to -400 cents.

That is, the pitch determination unit 8 determines the delay time j (pitch) that maximizes the autocorrelation σ' among the delay times j of the waveform shown in FIG. 5, for example.

Next, the post-processing section 9 receives the pitch output from the pitch determination section 8 and the stability output from the multiplication circuit 6, and outputs a final pitch. The operation of the post-processing section 9 will be explained in detail with reference to FIG.

First, the pitch input is delayed by a predetermined time by a delay circuit 91, and then the difference from the original signal is calculated by a subtraction circuit 92. The result is compared with a predetermined value THI in a comparison circuit 93, and if the output of the subtraction circuit 92 (that is, the difference between the delayed signal and the current signal) is larger than the predetermined value TH'l, the signal H is set, and if it is smaller, the signal H is set. , are output to the NAND circuit 95. The above configuration is for detecting a sudden change in pitch, and the signal H
This outputs the following.

On the other hand, the stability is compared with a predetermined value TR2 by a comparison circuit 94. Then, when the stability is higher than the predetermined value TH2, the signal H is output, and when it is lower, the signal H is inverted by the inverter 97 and output to the NAND circuit 95.
□ The Nil 1 circuit 95 includes the comparison circuits 93 and 9 as described above.
Receive the output of 4 and take its Nando. That is, when the pitch changes suddenly, the stability is referred to, and if the stability is high at that time, the pitch output is outputted to the outside via the AND circuit 96. Furthermore, if the stability is low when the pitch changes suddenly, the sudden change will be ignored.

From the above, the finally extracted pitch is output.

[Effects of the Invention] As explained above, according to the present invention, input sound! When extracting pitch from a signal in real time, high frequency,
A pitch extraction device is provided that can suppress small amplitude portions and further emphasize large amplitude, low frequency signals. Therefore, even when applied to musical sounds, etc., stable and smooth pitch extraction can be performed over a wide pitch range.

[Brief explanation of the drawing]

FIG. 1 is a schematic configuration diagram of a pitch extraction device according to the present invention, FIG. 2 is a block diagram showing the configuration of a pitch extraction device according to an embodiment of the invention, and FIG. FIG. 4 is a block diagram showing an example of the circuit of the noise level determination section of the extraction device. FIG. 4 is a block diagram showing an example of the circuit of the post-processing section of the pitch extraction device of FIG. 2. FIG. FIG. 6 is a graph showing an example of the result of calculating an autocorrelation function. 1: A/D converter, 2: Noise level judgment section, 3: Gate, 4: EC value calculation section, 5.6 Square multiplication circuit, F: Autocorrelation section, 8: Pitch judgment section, 9: Post-processing section . Patent Applicant Yamaha Corporation Representative Patent Attorney Tetsuya Ito

Claims (1)

[Claims] (1) A pitch extraction device equipped with pitch extraction means for extracting the pitch of an acoustic signal waveform, further comprising means for calculating stability showing a higher value for larger amplitudes and lower frequencies based on the acoustic signal waveform, A pitch extraction device characterized in that the pitch extraction means extracts the pitch based on the product signal output from the stability calculation means.