JPH0465398B2 - - Google Patents

Description

[Detailed description of the invention] [Technical field to which the invention pertains] The present invention speeds up the playback speed of recorded audio by q times (the length of the time axis is compressed to 1/q) or
When slowing down to q (the length of the time axis is expanded by q times), the instantaneous value of the waveform changes by q times or 1/q times (as a result, the resulting broadening of the audio spectrum is also
(becomes q times or 1/q times the original spectrum)
A time axis compression/expansion device that can change the playback speed without impairing auditory individuality and naturalness information by multiplying the instantaneous frequency signal information derived from the phase term of the analyzed audio signal by 1/q or q times. Regarding.

[Prior art and its problems] Generally, when the playback speed of a playback device that plays back recorded sound is increased, the playback of the entire recorded sound ends quickly. Furthermore, the amplitude envelope signal of the audio time waveform also changes slowly. Furthermore, the change in the instantaneous value of the waveform of the audio time signal, that is, the change in frequency with respect to time, becomes faster. Since the amplitude envelope signal of the speech time signal is redundant information for language judgment, it has little adverse effect on language judgment from an auditory perspective. However, if the instantaneous value of the waveform of the audio time signal changes quickly, the pitch will become higher audibly, the quality of the voice will change, and the individuality of the speaker will be lost. Conversely, when the playback speed is slowed down, a change in the instantaneous value of the waveform causes a similar change in the quality of the audio.

FIG. 5 explains this. That is, FIG. B shows a basic waveform diagram of a real-time signal (time-axis compressed signal) reproduced by compressing the time axis with respect to the original signal S(t) in FIG. A. These are composed of amplitude envelope signals α(t), α ₁ (t) and an alternating zero wave FM(t) (that is, a phase change of the waveform). This amplitude envelope signal α(t) changes relatively slowly with respect to the waveform frequency component of the original signal s(t). In other words, since it can be approximated that each frequency component is concentrated in a different band, the original signal s(t) can be approximated by the following equation (1).

S(t)=α(t)・FM(t)...(1) In this way, by considering qualitatively, the zero alternating wave FM(t) is a signal whose amplitude is constant and only the phase changes, i.e. It can be regarded as a phase modulated wave or FM wave. Considering this, the change in the instantaneous value of the reproduced waveform due to the condition of q times the reproduction speed is a change in the phase change by q times. Therefore, if only this phase change is multiplied by 1/q, it returns to the change in the instantaneous value of the original waveform. This is shown in FIG. 5C.

As explained above, for example, if a playback tape is
When the speed is doubled, the waveform of the reproduced signal becomes 1/q times as long in time. For this reason, the instantaneous phase change changes q times faster, that is, the instantaneous frequency also changes q times faster. Therefore, as mentioned above, if the phase or frequency level can be manipulated to return to that of the original signal, it is possible to calibrate the waveform while retaining the original characteristics of the speech, even in the case of rapid tongue twisters. It is.

The above method can be implemented for speech or other quasi-stationary time signals by using analytic signals. The analytic signal is a real-time audio signal S(t) and a Hilbert-transformed signal of this time signal S(t), that is, a signal s^(t) in which each frequency component is phase-shifted by 90 degrees, as expressed by the following equation (2). signal defined by
It is known as Sa(t).

Sa(t)=S(t)+jS^(t)...(2) Here, j is an imaginary unit. This equation (2) can be expressed as a complex exponential function as shown in the following equation (3).

Sa(t)=α(t)exp[jφ(t)]=α(t){Cos
[φ(t)]+jsin[φ(t)]}...(3) However, α(t)=√S ² (t)+S^ ² (t)...(4) φ(t)=tan ^{- 1} S^ ² (t)/S(t) (5) Therefore, the equation (4) is the envelope information of the analyzed audio signal, and the equation (5) is the phase information of the analyzed audio signal.

From equation (3), the real-time audio signal is α(t)Cos
Corresponds to [φ(t)].

From this, the information of the audio real-time signal is the envelope information α(t) and the phase information φ of the analyzed audio signal.
It can be seen that all are included in (t).

Proposals using such analytical signal methods have already been published. For example, one of the proposals (proposal 1) is Proceeding of the IEEE, Vol 55,
No.3, March, 1967, p396-401 MR
Schroeder, JLFlamagan, senior member,
“Bandwidth” by IEEE, and EALundry
Compression of Speech by Analytic−Signal
Proposal 1 discloses a processing system that uses the analytic signal method and Hilbert transform, and is limited to calibration especially for 1/2 playback speed.The other proposal (Proposal 2) is NHK Technical Research, 1972, No.
Volume 23, No. 6, Volume 127, P461-471, ``Compression and expansion of speech duration using analytic signal method'' by Yoshinori Totsuka. This proposal discloses a processing system that uses Proposal 1 to calibrate not only 1/2 speed but also double playback speed. Also, the phase change φ
It is only mentioned that by multiplying (t) by 1/q, it is possible to perform a calibration operation for changing the playback speed by a factor of q. Therefore, since Proposal 1 is included in Proposal 2, explaining Proposal 2, in the 1/2 and 2 times calibration operations, the phase change φ(t)
are 1/2φ(t) and 2φ(t), so FM(t), which is the FM wave part of the audio signal, is Cos[1/2φ
(t)], Cos[2φ(t)], and proposes an apparatus configuration that corresponds to the half-angle and double-angle formulas of the cosine function. Furthermore, a q-fold calibration operation is similarly disclosed for the phase change φ(t).
However, due to reasons such as considering the phase change φ(t) in the calibration operation, the following problems arose in realization.

(a) The sign is indeterminate because phase changes are processed.
Therefore, it becomes necessary to perform code discrimination processing called a switcher.

(b) Obtained using Hilbert transform and a set of transfer functions. Therefore, an adjustment operation using delay processing is necessary.

(c) Only 1/2x and 2x calibration operations are realized;
It is difficult to implement a general q-fold calibration operation.

[Purpose of the invention]

In view of the above-mentioned points, it is an object of the present invention to provide a time axis compression/expansion device that effectively solves the problems of the prior art and can cope with any change in playback speed without any difference.

[Key points of the invention]

In order to achieve such an objective, the inventors of the present invention have repeatedly conducted numerous studies and experiments, and as a result, have developed a method for calibrating the pitch change associated with a change in playback speed, e.g. Since the instantaneous frequency signal S(t) is a temporal change in phase information, equation (6) is derived from equation (5).

fs(t)=1/2π・dφ(t)/dt=1/2
π(dS^(t)/dtS(t)−s^(t)dS(t)/dt)
/(s ² (t) + s^ ² (t))
...(6) Focusing on this equation (6), for an audio signal whose playback speed is multiplied by q, this instantaneous frequency signal S
(t) is obtained from both the reproduced audio signal and the reproduced audio signal subjected to Hilbert transform processing, and further signal processing is performed to obtain instantaneous frequency signal information multiplied by 1/q, and the original real time is calculated from equation (6) of the analytic signal. Return the audio signal to instantaneous frequency signal information. The envelope information is synthesized with the playback speed changed by q times, and the original audio (sound at the time of recording) is
They reached the conclusion that it is possible to reproduce speech twice as fast without losing its individuality or phonology, and this was confirmed accurately through computer simulation.

In other words, when expanding the playback speed of recorded audio to q times the normal speed or compressing it to 1/q times,
The change in the instantaneous value of the reproduced waveform that changes by a factor of q or 1/q is an analytical audio signal defined from the audio time signal S(t) and the Hilbert-transformed signal S^(t) of this audio time signal. Phase information φ of Sa(t)
(t)=tan ^-1 Deriving the instantaneous frequency signal S(t) as the temporal change of S^(t)/S(t) as shown in equation (6),
This instantaneous frequency signal S(t) is multiplied by 1/q or q
It is characterized by performing multiplication signal processing, processing and synthesizing the waveform using equation (6) of the analytic signal, and calibrating and outputting changes in pitch due to changes in playback speed.

[Embodiments of the invention]

Next, embodiments of the present invention will be described in detail based on the drawings.

FIG. 1 is a basic configuration diagram of an embodiment of the present invention, and FIG. 2 is a block diagram of one channel in FIG. In FIGS. 1 and 2, the time axis compression/decompression device 1 mainly includes a Hilbert transform circuit C, a filter bank FB, an instantaneous frequency signal extraction circuit D consisting of four channels in this embodiment, and a reproduction speed change multiplier circuit E. , an FM wave output circuit F, an envelope signal output circuit G, a low-pass filter LPF, an envelope FM wave multiplier circuit H, a reproduced signal output section M, and the like.

Incidentally, reference numeral 2 shown in FIG. 2 is, for example, a tape recording/playback device, a disc recording/playback device, a memory reading device, or the like, which outputs an extracted signal to the time axis compression/expansion device 1. Further, A is installed in the above-mentioned playback device 2, and is a playback speed change that sequentially outputs a playback speed change q or 1/q to indicate how many times faster the playback is than the recording speed V (Fig. 1). It is an output device. B is a recorded information output device that outputs recorded information whose speed has been changed, mainly audio. The Hilbert transform circuit C outputs, for example, a Hilbert transform pair S(t) and S^(t) of speech to an instantaneous frequency signal extraction circuit D and an envelope output circuit G. The playback speed change multiplier circuit E multiplies the instantaneous frequency signal S(t) by 1/q times or q times by the playback speed change q or 1/q by the playback speed change output device A. Note that the FM wave output circuit F calculates and outputs the equation (3) in which the amplitude envelope signal α(t) is constant. Further, the envelope signal output circuit G calculates and outputs the envelope signal α(t) using equation (4). The envelope FM wave multiplier circuit H multiplies the output of the FM wave output circuit F by the information of the envelope signal α(t), and outputs a signal whose pitch change is calibrated.

Note that the playback speed change q at this time is not limited to an integral multiple, and any value can be selected.

Next, FIG. 3 shows a waveform diagram of various quantities extracted by the analytic signal method in FIG. 1. In the diagram A
is the original real-time audio signal S(t), B is the envelope signal α(t), C is the phase term Cos[φ(t)], and D is
FM component FM(t), E is the instantaneous frequency S(t),
This vertical frequency fluctuation is multiplied by q or 1/q depending on the reproduction speed.

Next, FIG. 4 shows a calibration waveform diagram for changes in reproduction speed according to FIG. 1. In the figure, Figure A is the original real-time audio signal S(t), and Figure B is the reproduced real-time audio signal S ₁ (t) when the reproduction time axis compression rate is 60%.
So, C is also 80%, D is 100%, and E is 120%. Both are time waveforms obtained by changing the reproduction speed of the original real-time audio signal S(t). Therefore, even though the playback time changes, the pitch period is calibrated to remain constant and the phoneme of the original speech is maintained.

In other words, the audio signal is divided into bands, for example by a bandpass filter, but the instantaneous frequency is extracted for each band (finding the peak frequency).
Therefore, the Hilbert transform is performed as an operation for this purpose, and the waveform is sequentially processed, not the processing for each section. Processed, it is possible to continuously change the total playback speed by any factor.

〔Effect of the invention〕

As explained above, according to the present invention, the playback speed is increased by q times the normal speed of the original audio or by 1/q.
When the speed is doubled, the change in the instantaneous value of the reproduced waveform that changes by a factor of q or 1/q is calculated using the analytical signal method.The instantaneous frequency signal is derived as a temporal change in phase change using the phase information of the analyzed audio signal. By performing signal processing that is multiplied by 1/q or q using a calculation formula, changes in phoneme due to changes in playback speed are calibrated, negative effects on auditory sensation are eliminated, and the problems of the conventional technology are effectively solved. This has the effect of being able to respond to changes in playback speed.

Although the embodiment of the present invention will mainly be described with respect to recorded audio, it is not limited to this, but can also be applied to reproduced sounds from video cassettes, arbitrary vibration sounds, etc., and arbitrary time signals. It's something you get.

[Brief explanation of the drawing]

Fig. 1 is a basic configuration diagram of an embodiment of the present invention, Fig. 2 is a block diagram for one channel in Fig. 1, and Fig. 3 is a waveform of various quantities extracted by the analytical signal method shown in Fig. 1. Figure 4 is a waveform diagram calibrated for changes in playback speed according to Figure 1. Figure 5 is a diagram of the basic waveforms of the original signal and reproduced signal, and a waveform diagram calibrated for changes in playback speed. It is. 1: Time axis compression/expansion device, 2: Playback device, C:
Hilbert transform circuit, D: Instantaneous frequency signal extraction circuit, E: Playback speed change multiplication circuit, F: FM wave output circuit, G: Envelope signal output circuit, H: Envelope/
FM wave multiplier circuit.

Claims (1)

[Claims] 1. A waveform that changes q times or 1/q times the normal speed when the playback speed of audio recorded on a recording or memory is made q times faster or 1/q times slower than the normal speed. The change in the instantaneous value of is expressed as the voice time signal s(t) and the Hilbert-transformed signal s^ of this voice time signal.
From the phase information φ(t)=tan ^-1 s^(t)/s(t) of the analytic audio signal sa(t) defined from (t), the instantaneous frequency signal fs(t) of the analytic audio signal is the temporal change in the phase information, and the following formula fs(t)=1/2π・dφ(t)/dt=1/2π(d
s^(t)/dts(t)−s^(t)ds(t)/dt)/(s ²
(t)+s^ ² (t)), and the instantaneous frequency signal fs(t) is 1/
A time axis compression/expansion device characterized in that it performs signal processing that multiplies by q or q to calibrate and output changes in pitch due to changes in playback speed.