Method For Decorrelating A Set Of Simulated Audio Signals

Abstract

A computer-implemented method for decorrelating a set of simulated audio signals in a virtual environment is disclosed. The method comprises the steps of: obtaining a first set of decorrelation filters; evaluating the first set of decorrelation filters to determine a similarity metric representing the similarity between the first set of decorrelation filters; if the similarity metric is above a predetermined threshold, regenerating the set of decorrelation filters so as to reduce the similarity metric; and applying the regenerated set of decorrelation filters to the simulated audio signals.

Claims

1. A computer-implemented method for decorrelating a set of simulated audio signals in a virtual environment, the method comprising: obtaining a set of decorrelation filters; evaluating the set of decorrelation filters to determine a similarity metric representing a similarity among the set of decorrelation filters; determining that the similarity metric is above a predetermined threshold; regenerating the set of decorrelation filters to reduce the similarity metric; and applying the regenerated set of decorrelation filters to the simulated audio signals.

2. The computer-implemented method according to claim 1, wherein determining the similarity metric comprises calculating a cross-correlation between each decorrelation filter within the set of decorrelation filters.

3. The computer-implemented method according to claim 1, wherein the set of decorrelation filters are iteratively regenerated until the similarity metric is below the predetermined threshold.

4. The computer-implemented method according to claim 1, wherein the set of decorrelation filters alter phases of the audio signals in the set of simulated audio signals.

5. The computer-implemented method according to claim 1, wherein the set of decorrelation filters comprises a set of all-pass filters.

6. The computer-implemented method according to claim 1, wherein the set of decorrelation filters comprises: a frequency selecting variable that selects frequencies in the set of simulated audio signals to be filtered; and a phase delay variable that determines a maximum phase that the selected frequencies are to be delayed.

7. The computer-implemented method according to claim 1, wherein the set of decorrelation filters is obtained by randomly generating the set of decorrelation filters.

8. The computer-implemented method according to claim 7, wherein randomly generating the set of decorrelation filters comprises randomly generating a frequency selecting variable and a phase delay variable.

9. The computer-implemented method according to claim 1, wherein the set of simulated audio signals comprises multiple copies of an audio signal.

10. The computer-implemented method according to claim 9, wherein the multiple copies constructively, destructively, or both constructively and destructively interfere with one another.

11. The computer-implemented method according to claim 1, further comprising generating the set of audio signals according to acoustic parameters of the virtual environment.

12. The computer-implemented method according to claim 1, wherein regenerating the set of decorrelation filters further comprises applying at least one of an optimisation algorithm or machine learning technique.

Description

BRIEF DESCRIPTION OF DRAWINGS

[0017] Embodiments of the invention are now described, by way of example, by reference to the drawings, in which:

[0018] FIG. 1 is a flow diagram of a computer-implemented method in an embodiment of the invention;

[0019] FIG. 2 is a flow diagram of a computer-implemented method in another embodiment of the invention; and

[0020] FIG. 3 is a table comprising a randomly generated set of decorrelation filters in an embodiment of the invention.

DETAILED DESCRIPTION

[0021] FIG. 1 is a flow diagram of a computer-implemented method 100 for decorrelating a set of simulated audio signals in a virtual environment in an embodiment of the invention. The method 100 comprises the steps 102, 104, 106, 108 and 110.

[0022] Step 102 comprises obtaining a set of decorrelation filters. Step 104 comprises evaluating the obtained set of decorrelation filters to determine a similarity metric. The similarity metric indicates the similarity between each of the decorrelation filers in the obtained set of decorrelation filters. Step 106 comprises determining whether the similarity metric is above a predetermined threshold or not. If the similarity metric is above the predetermined threshold, the decorrelation filters in the set are deemed to be too similar to one another and are not acceptable for application to the set of simulated audio signals. The set must therefore be regenerated at step 108 so as to reduce the similarity metric. The regenerated set of decorrelation filters can then be applied to the set of simulated audio signals at step 110 to decorrelate the set of simulated audio signals.

[0023] If the similarity metric is determined to be below the predetermined threshold at step 106, the decorrelation filters in the set are deemed to be dissimilar enough to one another and are acceptable for application to the set of simulated audio signals at step 110 to decorrelate the set of simulated audio signals. In this scenario, it is not necessary to regenerate the set of decorrelation filters at step 108.

[0024] FIG. 2 is a flow diagram of a computer-implemented method 200 for decorrelating a set of simulated audio signals in a virtual environment in another embodiment of the invention. The method 200 comprises the steps 201, 202, 204, 206, 208 and 210, however further steps may be additionally included at various stages in the method.

[0025] Step 201 comprises generating a set of simulated audio signals according to acoustic parameters of a virtual environment. Step 202 comprises randomly generating a set of decorrelation filters which each include a frequency selecting variable and a phase delay variable. The set of decorrelation filters is evaluated to determine a similarity metric at step 204. The similarity metric indicates the similarity between each of the decorrelation filters, including their frequency selecting variables and phase delay variables, in the randomly generated set of decorrelation filters. At step 206 it is determined whether the similarity metric is above or below a predetermined threshold. If the similarity metric is determined to be below the predetermined threshold, the randomly generated set of decorrelation filters are deemed to be dissimilar enough to one another and are acceptable for application to the generated set of simulated audio signals at step 210 to decorrelate the set of simulated audio signals thereby reducing phasing artefacts in the audio output of the virtual environment.

[0026] If the similarity metric is determined to be above the predetermined threshold, the randomly generated decorrelation filters are deemed to be too similar to one another and are therefore not acceptable for application to the set of simulated audio signals. If this is found to be the case, a new set of decorrelation filters is randomly generated at step 208. The method then returns to step 204, where the regenerated decorrelation filters are evaluated to determine a similarity metric associated with the regenerated decorrelation filters. The method then progresses to step 206, where it is determined whether the similarity metric associated with the regenerated decorrelation filters is above or below the predetermined threshold. This regeneration process is iterated over until the regenerated decorrelation filters are determined to have a similarity metric that is below the predetermined threshold. The regenerated set of decorrelation filters are then acceptable for application to the generated set of simulated audio signals at step 210 to decorrelate the set of simulated audio signals thereby reducing phasing artefacts in the audio output of the virtual environment.

[0027] In this embodiment, step 201 comprises generating a set of simulated audio signals according to acoustic parameters of a virtual environment. The set of simulated audio signals which represent sounds comprises multiple copies of the same audio signal. This can occur in a video game environment, for example, when an audio signal reflects off multiple simulated surfaces in the video game environment producing multiple delayed copies of the original audio signal. The multiple copies of the original simulated audio signal then interfere with each other, either constructively and/or destructively, producing phasing artefacts in the audio output of the video game environment.

[0028] Step 202 comprises randomly generating a set of decorrelation filters. In this embodiment, each decorrelation filter in the randomly generated set includes a frequency selecting variable and a phase delay variable. When the set of decorrelation filters is randomly generated, the frequency selecting variable and the phase delay variable are each randomly generated for each decorrelation filter in the set. The frequency selecting variable indicates the frequency or the frequencies for which the decorrelation filter is to apply. The phase delay variable determines to what degree the filter is to induce a frequency delay on the frequency or the frequencies indicated by the frequency selecting variable.

[0029] For example, as shown in FIG. 3, a first randomly generated decorrelation filter F1 in the set could have a frequency selecting variable of 500 Hz to 1000 Hz and a phase delay variable of 3?/4. If applied to the simulated audio signals, this filter would introduce a random phase shift between 0 and 3?/4 to randomly selected frequencies within the 500 Hz to 1000 Hz range.

[0030] A second randomly generated decorrelation filter F2 in the set could have a frequency selecting variable of 9000 Hz to 15000 Hz and a phase delay variable of ?/6. If applied to the simulated audio signals, this filter would introduce a random phase shift between 0 and ?/6 to randomly selected frequencies within the 9000 Hz to 15000 Hz range.

[0031] In the embodiment of FIG. 2, the randomly generated set of decorrelation filters comprises a plurality of randomly generated all-pass filters that each differently alter the phases of different frequencies in the set of simulated audio signals, without altering the amplitudes of the simulated audio signals.

[0032] In alternative embodiments, the randomly generated set of decorrelation filters may comprise other types of filters that are randomly generated. For example, the randomly generated set of decorrelation filters may comprise high pass, low pass and/or band pass filters that are each randomly generated. Alternatively, any all-pass filter generation method may be used to provide the randomly generated set of decorrelation filters.

[0033] At step 204, the randomly generated set of decorrelation filters are evaluated to determine a similarity metric. The similarity metric indicates the similarity between each of the decorrelation filters, specifically the similarities between the frequency selecting variables and the phase delay variables of each of the decorrelation filters, in the randomly generated set of decorrelation filters. In this embodiment, the similarity metric is calculated by measuring the cross-correlation between the frequency selecting variables and the phase delay variables of each of the decorrelation filters in the randomly generated set.

[0034] For example, with reference to FIG. 3, the correlation between the phase delay variables of each of the filters F1-F5 could be calculated along with the cross-correlation between the frequency selecting variables of each of the filters F1-F5, or vice versa, to determine an overall similarity metric. In alternative embodiments, the similarity metric may be calculated using only a cross-correlation metric.

[0035] At step 206, it is determined whether the similarity metric is above or below a predetermined threshold. In this embodiment, the predetermined threshold is a fixed value. If the calculated similarity metric is of a greater value than that of the predetermined threshold, then the randomly generated decorrelation filters are deemed to be too similar to one another to be applied to the simulated audio signals. If the calculated similarity metric is of a lesser value than that of the predetermined threshold, then the randomly generated decorrelation filters are deemed to be dissimilar enough from one another and are therefore suitable to be applied to the simulated audio signals. In alternative embodiments, the predetermined threshold may not be a fixed value. For example, the predetermined threshold may be programmed to change over time.

[0036] If the calculated similarity metric is of a greater value than that of the predetermined threshold, and therefore that the randomly generated decorrelation filters are too similar to one another to be applied to the simulated audio signals, then a set or subset of decorrelation filters is regenerated at step 208. The regenerated set of decorrelation filters is generated in the same way as that of the randomly generated set of decorrelation filters as described in step 202.

[0037] The regenerated set of decorrelation filters is analysed according to steps 204 and 206 described above. The generated sets of decorrelation filters are iteratively regenerated until a regenerated set of decorrelation filters has an associated similarity metric that is below the predetermined threshold, as shown in FIG. 2.

[0038] At step 210, once it has been determined that the similarity metric of the set of decorrelation filters is below the predetermined threshold and that each of the filters are suitably dissimilar from one another, the filters are applied to the simulated audio signals to minimise the phasing artefacts in the video game environment. The decorrelation filters introduce a phase delay for select frequencies in the simulated audio signals, thereby reducing the effects of interference between multiple copies of the simulated audio signals. The decorrelation filters also pass the select frequencies equally in gain to prevent the multiple copies of the simulated audio signals from being overly distorted.

[0039] In alternative embodiments, the methods discussed herein may further comprise the step of applying optimisation algorithms or machine learning techniques to generate a set of decorrelation filters as opposed to randomly generating a set of decorrelation filters. More specifically, machine learning techniques may be applied to generate a set of all-pass filters. The decorrelation capabilities of each all-pass filter in the set and the dissimilarities between each all-pass filter in the set are maximised in a single optimisation process.