JP4845407B2 - How to generate a reference filter - Google Patents

Description

  The present invention relates to a method for correcting a filter generated from sound collected using a dummy head in a reference environment.

  Generally, sound output from a speaker is reflected or attenuated in a reproduction environment in a complicated manner, and then reaches a sound receiving point. For this reason, even when a reference sound having a flat characteristic (a sound having an equivalent sound pressure (dB) in a predetermined range of frequencies) is recorded, the degree of attenuation differs depending on the frequency, and the delay is also different. Occurs. This can be quantitatively expressed as a peak (crest) or dip (valley) of sound pressure (dB) or a waveform delay (shift) in the case of frequency analysis. Such a change in amplitude or delay according to the frequency (band) is referred to as a frequency characteristic in the environment.

  Conventionally, using this frequency characteristic, for example, a reference sound is collected in a reference environment such as a concert hall, a filter that reproduces the frequency characteristic is generated, and a sound source is reproduced through this filter. The sound effect is as if it was heard in a concert hall. Hereinafter, the sound in such a specific (preferred) environment (sound field) is referred to as a reference sound, and a filter for reproducing the frequency characteristics is referred to as a reference filter.

  As a reference filter, an equalizer using an electronic circuit (hardware) has been known for a long time. The digital filters (software) that have been remarkably developed in recent years are FIR filters (Finite Impulse Response Filter) with high reproducibility but high computational load, and IIR filters (Infinite-duration Impulse Response Filter) with low reproducibility but low computational load It is roughly divided into

  By the way, humans have two ears and always sense the position of the sound source. Therefore, in order to reproduce the sense of distance to the sound source and how the sound spreads, it is necessary to investigate how it is heard by two ears. It is possible to use two microphones at this time, but not only the delay due to the position (difference in arrival time from the sound source to the position of the left and right ears), but also the directivity and sound that both ears are open to the outside In addition, it is necessary to consider the effects of sneaking around the nose and the ear lobe, head, and ear canal. Therefore, sampling is generally performed by installing a microphone at the eardrum position of both ears using a doll called a dummy head (head torso) that faithfully reproduces the ear part.

  However, when the reference filter is generated based on the recording data using the dummy head, the frequency characteristic depending on the dummy head becomes more dominant than the frequency characteristic depending on the environment. Deterioration may be felt. If the sound source filter is adjusted so that it sounds flat for the dummy head, and a person listens to it as it is, the frequency characteristic depending on the dummy head and the frequency characteristic depending on the person overlap, and the filter is applied too much. For example, if the dummy head is assumed to have significant bass attenuation, and the sound source is played back using a reference filter that amplifies the bass range to correct this, the listener will have a low-pitched ear. As a result, the bass becomes too loud.

In Patent Document 1 (Japanese Patent Laid-Open No. 6-147968), an equalizer having an inverse characteristic of a frequency characteristic diffracted wave of a dummy head (head torso simulator) is prepared, recorded from the dummy head via the equalizer, and recorded on the reproduction side. A configuration for optimizing an equalizer (filter) is described. Patent Document 1 states that the reproduction-side equalizer thus obtained is an optimum equalizer for obtaining a balanced frequency characteristic in terms of audibility. Note that Patent Document 1 aims to flatten the way in which sound is heard in a specific environment (in a vehicle), and is not the purpose of generating a reference filter, but it is a common element that eliminates the effects unique to the dummy head. have.
JP-A-6-147968, 0016, 0023-0025, FIG.

  As described above, the development of digital filters has been remarkable in recent years, and it is possible to use FIR filters with high reproducibility. By using the FIR filter, the reproducibility of frequency characteristics such as a concert hall can be enhanced, and the quality and the degree of freedom of sound quality correction can be greatly improved. Further, by using the FIR filter, it is possible to cancel crosstalk (CTC: for example, canceling the sound coming out of the right speaker and entering the left ear by emitting the sound in the opposite phase from the left speaker).

  However, if an FIR filter is used to generate a filter with an inverse characteristic to avoid the frequency characteristic unique to the dummy head, the FIR filter is highly reproducible, so the influence of the dummy head and the environment-dependent frequency characteristic are also eliminated. Will end up.

  That is, while enjoying the benefits of high reproducibility using the FIR filter, the effects dependent on the outside of the dummy head, such as the position of both ears, the directivity facing outward, the diffraction that wraps around the face and head, are reproduced, In addition, there is a demand to eliminate influences depending on the inside of the dummy head, such as reflection in the ear canal, reflection in the head, and resonance.

  In Patent Document 1, in order to eliminate the influence unique to the dummy head, the frequency characteristics of the dummy head are obtained by recording in an ideal free sound field having no reverberation or reverberation (Patent) Reference 1, paragraphs 0009 and 0011). However, it is difficult to construct such an ideal environment, and it is also difficult to reproduce a sound source such as a hall in an environment close to this. Furthermore, since the way of sound transmission differs between such a free sound field and a specific environment such as a hall, the influence generated from within the dummy head should also be different. There is a possibility that the influence of the dummy head cannot be completely eliminated.

  SUMMARY OF THE INVENTION An object of the present invention is to provide a method for generating a reference filter that eliminates the effects unique to a dummy head and improves sound quality while maintaining a sense of presence and reverberation. At the same time, it is an object to provide a generation method capable of improving localization, changing the sense of presence and localization.

  In order to solve the above-described problem, a representative configuration of a reference filter generation method according to the present invention calculates a first frequency characteristic from sound data acquired using a dummy head, and the first frequency characteristic is A logarithmic averaging process is performed on the amplitude at a predetermined interval to calculate a second frequency characteristic, a temporary filter having a characteristic opposite to the logarithmic averaging second frequency characteristic is generated, and the temporary filter The third frequency characteristic corrected by multiplying the first frequency characteristic and the first frequency characteristic is calculated, and a filter is generated based on the third frequency characteristic.

  The filter is preferably an FIR filter.

  Preferably, the reference filter includes left and right filters generated based on sound data acquired from at least a dummy head, and amplification levels of the left and right filters are set independently on the left and right.

  It is preferable that the reference filter includes left and right filters generated based on sound data acquired from at least a dummy head, and sets a delay time between the left filter and the right filter independently. In particular, the left and right levels and the delay time are preferably matched.

  Sounds emitted from at least a pair of left and right sound sources are acquired from the left and right ears of the dummy head, and a crosstalk filter is generated based on sound data that has reached the other ear from the left and right sound sources, and the other side Preferably, a filter is generated based on sound data that has reached the other ear from the sound source and the crosstalk filter is added to the filter. .

  According to the present invention, it is possible to obtain a reference filter that eliminates the influence of the dummy head and improves the sound quality while maintaining a sense of presence and reverberation. Also, the localization of the sound source can be stabilized by making it possible to set the left and right levels and the delay separately. In addition, by setting the left and right crosstalk levels to the left and right separately, the presence and localization can be made variable.

  An embodiment of a reference filter generation method according to the present invention will be described. In addition, in this specification and drawing, about the component which has the substantially same function structure, duplication description is abbreviate | omitted by attaching | subjecting the same code | symbol.

  FIG. 1 is a schematic configuration diagram illustrating a system for generating a reference filter in the present embodiment. It is assumed that the system is installed in a specific environment for creating a reference filter, such as a concert hall.

  As shown in FIG. 1A, a left speaker L and a right speaker R are installed as sound sources. A dummy head 1 is installed at a position facing the sound source. Microphones 2 </ b> L and 2 </ b> R are attached to the left and right eardrum positions of the dummy head 1 and connected to the recording unit 3. A computing unit 4 is connected to the recording unit, and further a recording unit 5 for recording data and computation results, an output unit 6 for outputting these on a sheet or screen, and various operations and data input. An operation input unit 7 is provided.

  Recording (sampling) is performed by outputting a pulse wave from only one of the left and right speakers L and R and recording from the left and right microphones 2L and 2R of the dummy head 1, respectively. As shown in FIG. 1B, a pattern output from the left speaker L and input to the left microphone 2L is C11, a pattern output from the left speaker L and input to the right microphone 2R is C12, A pattern output from the speaker R and input to the left microphone 2L is C21, and a pattern output from the right speaker R and input to the right microphone 2R is C22.

  In this embodiment, a reference filter for reproducing the frequency characteristics of the specific environment is created from the sampling of the four patterns. In the following description, a detailed description of a general filter generation method and a crosstalk filter generation method will be omitted, and a correction method that is a feature of the present invention will be mainly described.

[Correction to eliminate the influence of dummy head]
First, correction for eliminating the influence of the dummy head will be described with reference to FIG. This correction is performed for sound data acquired in the four patterns C11, C12, C21, and C22. Therefore, in the present embodiment, only the case of the C11 pattern will be described with reference to FIG.

  FIG. 2A shows the waveform of the reference sound to be output. The reference sound outputs a sound having a flat frequency characteristic for a short time, and is a so-called impulse response sampling.

  FIG. 2B shows an example of sound data when this is recorded. The graph of time and amplitude shows how the sound is transmitted, that is, how the sound reaches and extends due to reverberation (reflection) and attenuates. The graph of frequency and sound pressure (dB) shows the first frequency characteristic (Impulse Response), and it can be seen that there is a sound that reaches well with the frequency (peak) and a sound that attenuates (dip). The frequency characteristics can be calculated by Fourier transforming the impulse response waveform.

  By the way, since the sound has a wavelength, the fluctuation becomes fine when the reflection position is far (the delay time is long), and the fluctuation becomes gentle when the reflection position is close (the delay time is short). That is, referring to the graph of frequency and sound pressure in FIG. 2B, a fine wave and a large wave can be felt, but the fine wave is mainly due to reflection from a far wall (depending on the environment). It can be understood that the large wave is mainly due to the influence in the dummy head. Therefore, if a large fluctuation is eliminated while leaving a fine fluctuation, the influence unique to the dummy head can be eliminated while maintaining the presence and reverberation depending on the environment. In practice, the frequency characteristics of speakers and microphones should be considered, but in this embodiment, they are omitted for the sake of simplicity.

  Therefore, as shown in FIG. 2C, the logarithmic averaging process is performed on the frequency characteristics at a predetermined interval to calculate a second frequency characteristic (Impulse Response Average). The logarithmic average is the average of the values obtained by taking the natural logarithm. Here, averaging is performed only on the amplitude, and the phase information remains. This is because if the phase information is also averaged, the waveform is significantly attenuated.

  The predetermined interval is a value to be adjusted as appropriate. That is, as the interval is increased, the logarithmically averaged waveform changes more slowly, and depends on only the influence in the dummy head, leaving an environment-dependent waveform. However, if it is too loose, the influence in the dummy head cannot be reproduced, and the original function and effect cannot be achieved. Therefore, it is necessary to set appropriately in consideration of the sensitivity of the system and the accuracy of the target reference filter.

  Similarly, as shown in FIG. 2C, an inverse filter (temporary filter) is generated by reversing the phase of the logarithm averaged waveform (Impulse Response Average). By multiplying (convolution integration) this inverse filter and the first frequency characteristic obtained first, the first frequency characteristic is corrected and a third frequency characteristic is generated.

  FIG. 2D shows the waveform and frequency characteristics of the corrected sound data. Here, referring to the graph of frequency and sound pressure, it can be seen that there is no large variation in the third frequency characteristic, and the overall characteristic is flat. Minor fluctuations indicate fluctuations depending on the environment. A result of Fourier transform of this frequency characteristic is shown in a graph of time and amplitude.

  Now, when the reference sound as shown in FIG. 2 (a) is to be emitted and converted to emit the sound as shown in the time and amplitude graph of FIG. 2 (d), for example, with headphones. Even when listening, you can reproduce the feeling as if you were listening in a concert hall. Therefore, based on the third frequency characteristic, a filter that reproduces this is generated by an FIR filter and reproduced through this reference filter, whereby the above-described acoustic effect can be obtained.

  Here, by using the FIR filter, a reference filter having extremely fine reproducibility can be obtained. As described above, by making the overall frequency characteristic flat as preprocessing for generating the FIR filter, the reference filter which eliminates the influence of the dummy head and improves the sound quality while maintaining the sense of reality and reverberation, can do.

  In addition, according to the above generation method, it is not necessary to perform sampling separately by installing the dummy head in a special anechoic space or the like in order to eliminate the influence unique to the dummy head, and install it in a specific environment where a reference filter is desired. And sampling. Therefore, the labor of the sampling operation is reduced and the influence can be eliminated by using the sound that reaches the dummy head in the actual environment, so that the influence specific to the dummy head can be more appropriately eliminated.

[Amplification level and delay time correction]
Next, the correction of the amplification level and the delay time will be described with reference to FIG. This correction corrects each amplification level and delay time for the sound data acquired in the four patterns C11, C12, C21, and C22.

  As shown in FIG. 3, the reference filter 8 includes filters R11, R12, R21, and R22 corresponding to C11, C12, C21, and C22, respectively. Each filter is configured by an FIR filter.

  When sampling the reference sound using the dummy head 1, depending on the positional relationship between the speakers L and R and the measurement position and the state of reflection from the wall, as shown in the column of “impulse response” in FIG. There may be a difference between the 2L and 2R signal levels and the delay time. Although it is conceivable to re-take them until they become uniform, there is a problem that the burden of the sampling work increases.

  If the reference filter 8 is generated with a difference between the signal level and the delay time, the same difference is also reproduced in the sound reproduced later using this filter. Then, the place where the sound source localization is to be set in front of the listener is heard with a deviation.

  Therefore, in the present embodiment, when the first peak in the positive direction in the amplitude waveform is referred to as the first path, correction is performed so that the amplification levels of all the first paths C11 to C22 coincide. Further, correction is performed so that the start times of the first pass of C11 and C22 coincide. Here, the delay time d1 of C12 with respect to C11 and the delay time d2 of C21 with respect to C22 are maintained as they are. And based on the impulse response after correction | amendment, each filter R11-R22 of the reference filter 8 is produced | generated. That is, each of the filters R11 to R22 sets the amplification level and the delay time independently on the left and right.

  As a specific method, for example, a method of generating the filters R11 to R22 after correcting the acquired sound data in advance can be considered. That is, first, the gains are adjusted to match the sound data levels based on the amplitude waveform data, and the delays are corrected, and then the filters R11 to R22 are generated. As another method, it is possible to generate each filter without changing the level and delay of the sound data for the time being, and then correct the amplification level and delay time of the filter.

  By configuring as described above, the sound source can be positioned in front of the listener, so that the listener can hear the sound with a sense of stability. Further, since the configuration is such that the delay time of the crosstalk is maintained, it is possible to maintain a sense of presence (a feeling of spread).

[Correction of crosstalk level and delay time]
Next, correction of the crosstalk level and the delay time will be described with reference to FIG. As shown in FIG. 4, among the filters R11, R12, R21, and R22, the filter R12 generated from the sound data of C12 output from the left speaker L and input to the right microphone 2R, and the right speaker R The filter R21 generated from the sound data of C21 output from and input to the left microphone 2L is a crosstalk filter.

  To explain the outline of crosstalk, the sound that reaches the ear on the other side (for example, right) from the sound source on one side (for example, left) is called crosstalk. The filter used for this crosstalk is referred to as a crosstalk filter in this embodiment. That is, the crosstalk filter is filters R12 and R21 generated based on sound data acquired between a diagonal speaker and a microphone.

  In the description using FIG. 3, the levels of the crosstalk filters R12 and R21 are the same as the filters R11 and R22, and the delay time is maintained. However, different acoustic effects can be obtained by further changing the crosstalk level and the delay time. Specifically, if the level of crosstalk is increased, the sense of presence increases, and if the level is decreased, the localization is improved. Further, the position of localization can be adjusted by adjusting the delay time of the left and right crosstalk.

  Therefore, in this embodiment, the ratio level (gain) and delay time when adding the crosstalk filter can be set independently for the left and right. In FIG. 4, a block G represents gain adjusting means, and a block d represents delay adjusting means.

  By configuring as described above, it is possible to generate a reference filter that can also set the presence and the localization. The setting of the level of the crosstalk filter and the delay time may be performed when the reference filter is generated, but may be configured so that the parameters can be dynamically changed during reproduction.

  As mentioned above, although preferred embodiment of this invention was described referring an accompanying drawing, it cannot be overemphasized that this invention is not limited to the example which concerns. It will be apparent to those skilled in the art that various changes and modifications can be made within the scope of the claims, and these are naturally within the technical scope of the present invention. Understood.

  The present invention can be used as a method for generating a reference filter from speech collected using a dummy head in a reference environment.

It is a schematic block diagram explaining the system for producing | generating a reference filter. It is a figure explaining the correction method which eliminates the influence of a dummy head. It is a figure explaining the correction of an amplification level and delay time. It is a figure explaining the correction | amendment of the level of a crosstalk, and delay time.

Explanation of symbols

L ... Left speaker R ... Right speaker 1 ... Dummy head 2 ... Microphone 3 ... Recording unit 4 ... Calculation unit 5 ... Recording unit 6 ... Output unit 7 ... Operation input unit 8 ... Reference filter

Claims (5)

Obtaining sound data emitted from a sound source with a microphone provided in the dummy head , calculating a first frequency characteristic serving as a transfer characteristic from the sound source to the microphone ,
The only Oite amplitude calculating a second frequency characteristic by performing a logarithmic averaging process at predetermined intervals for the frequency to the first frequency characteristic,
Generating a temporary filter having a reverse characteristic by inverting the phase of the logarithm-averaged second frequency characteristic;
Calculating a third frequency characteristic Ri by the multiplying the said and the one o'clock filter first frequency characteristic,
A reference filter generation method, wherein a filter coefficient is generated from the third frequency characteristic. The reference filter generation method according to claim 1, wherein the filter using the filter coefficient is an FIR filter. The reference filter is provided with a left and right filter generated based on the sound data acquired by the microphone provided in at least the dummy head,
2. The reference filter generation method according to claim 1, wherein amplification levels of the left filter and the right filter are set independently on the left and right. The reference filter is provided with a left and right filter generated based on the sound data acquired by the microphone provided in at least the dummy head,
The reference filter generation method according to claim 1, wherein a delay time between the left filter and the right filter is set independently on the left and right. At least sounds emitted from the pair of left and right sound sources are acquired by microphones provided on the left and right ears of the dummy head, and crossed based on sound data that has reached the microphones of the other ear of the dummy head from the left and right sound sources. Generate a talk filter,
A filter is generated based on sound data that has reached the microphone on the other side of the dummy head from the sound source on the other side,
Method of generating a reference filter of claim 1, wherein the output signal of the filter when adding the output signal of the crosstalk filter can be set the rate for adding the left and right independently.

Images