122 and ears 124 of the user 120 transform audio output 128 generated by the device 100. Using the HRIR to render audio allows the audio output device to control the location 134 at which the user 120 would perceive the audio output 128. Alternatively or additionally, collectively, the left ear samples 202-1 and the right ear samples 202-2 can comprise a head-related transfer function (HRTF). Each HRTF is a function that indicates how the head 122 and ears 124 of a user 120 modify various audio frequencies before the audio is perceived by the user 120, and therefore how the head 122 and ears 124 of the user 120 transform audio output 128 of various frequencies generated by the device 100. Using the HRTF to render audio allows the audio output device to control the location 134 at which the user 120 would perceive the audio output 128. Each candidate audio profile 112 can correspond to and/or be based on one or more users 120 having a left ear 124-1 and/or a right ear 124-2 of a particular shape or size, wherein users 120 having ears 124 of similar shapes and sizes are likely to perceive audio output 128 rendered using a same candidate audio profile 112 as having originated from a similar location 134.

The audio profile determining engine 114 generates a vector representation 210 of one or more of the candidate audio profiles 112. As shown, the audio profile determining engine 114 performs an averaging 204 of the left ear samples 202-1 of the first candidate audio profile 112-1 to generate a left ear average sample 206-1, and also an averaging 204 of the right ear samples 202-2 of the first candidate audio profile 112-1 to generate a right ear average sample 206-2. The left ear average sample 206-1 can represent an average HRIR and/or an average HRTF of the left ear samples 202-1 of the candidate audio profile 112-1 (e.g., the impulse response and/or frequency response of the left ear 124-1 of a user 120 to all audio cues and/or audio frequencies). The right ear average sample 206-2 can represent an average HRIR and/or an average HRTF of the right ear samples 202-2 of the candidate audio profile 112-1 (e.g., the impulse response and/or frequency response of the right ear 124-2 of a user 120 to all audio cues and/or audio frequencies). The audio profile determining engine 114 performs a concatenating 208 of the left ear average sample 206-1 and the right ear average sample 206-2 to generate a first vector representation 210-1 of the first candidate audio profile 112-1. The vector representation 210 of each candidate audio profile 112 includes a response of the left ear 124-1 of the user 120 to one or more frequencies within a frequency range, such as the audible frequency range (e.g., 20 hertz to 20 kilohertz). While not shown, the audio profile determining engine 114 performs similar operations to generate vector representations 210 of each of the other candidate audio profiles 112 of the plurality of candidate audio profiles 112. In various embodiments, the vector representations 210 are compact and efficient representations of the corresponding candidate audio profiles 112. For example, a set of 312 left-ear measurements and a set of 312 right-ear measurements can be compactly represented as a single vector representation 210.

As shown, the audio profile determining engine 114 generates a matrix 212 of the vector representations 210 for each of the candidate audio profiles 112. In various embodiments, the audio profile determining engine 114 concatenates the vector representations 210 along a second axis to generate a two-dimensional matrix 212 of vector representations 210. Each vector representation 210 can be included as a column of the matrix 212.

As shown, the audio profile determining engine 114 performs a binning and normalization operation 214 to the matrix 212. In various embodiments, the audio profile determining engine 114 generates one or more bins, each representing a frequency range within a frequency spectrum of the matrix 212. In various embodiments, the bins can cover only a portion of the audible frequency spectrum (e.g., 1 kilohertz to 14 kilohertz), and other frequencies that are above or below the portion of the audible frequency spectrum can be discarded. The one or more bins can be of same, similar, and/or different sizes. The one or more bins can be spaced linearly or logarithmically over the frequency range. The audio profile determining engine 114 can aggregate the vector representations 210 comprising the columns of the matrix 212 into the bins. For example, for each vector representation 210-1 or column of the matrix 212, the audio profile determining engine 114 can determine an average of two or more vector elements representing audio samples of audio frequencies that are within the frequency range of one bin. Additionally, in various embodiments, the audio profile determining engine 114 normalizes the matrix 212. For example, for each vector element of each vector representation 210-1 or column of the matrix 212, the audio profile determining engine 114 can calculate a logarithmic value of the vector element, such as a normalized logarithmic intensity of a frequency response for each frequency bin within a binned human-audible frequency range. Alternatively or additionally, the audio profile determining engine 114 can normalize the matrix 212 in other ways, such as adding a positive or negative offset or bias to the vector element and/or clipping the vector element based on a high or low clipping value. Based on the binning and normalization operation 214, the audio profile determining engine 114 outputs a binned and normalized matrix 216.

As shown, the audio profile determining engine 114 performs a principal component analysis 218 of the binned and normalized matrix 216. In various embodiments, the audio profile determining engine 114 determines, among a feature set of the binned and normalized matrix 216, a reduced feature set of features that are representative of the matrix 212. That is, the audio profile determining engine 114 determines, among the feature set of the binned and normalized matrix 216, an excludable feature set of features that are not representative of the matrix 212. The audio profile determining engine 114 can retain the reduced feature set and exclude the excluded feature set of the binned and normalized matrix 216 to generate a reduced matrix 220. In various embodiments, the principal component analysis reduces a dimensionality of each vector representation 210 of the matrix 212 from 13,000 features (e.g., 13,000 frequency bins) to 8 features (e.g., 8 frequency bins). The reduced matrix 220 efficiently represents the matrix 212 of vector representations 210 of the candidate audio profiles 112 in a manner that retains significant features in a binned and normalized manner, while removing other features that are not representative of the matrix 212 and the candidate audio profiles 112 encoded into the matrix 212. The reduced matrix significantly reduces the computing cost of determining an audio profile to be used for the device 100 from among the candidate audio profiles. The reduced matrix also allows the selection steps to focus on the most significant differences in the audio features of candidate audio profiles, such as the audio features that distinguish the candidate audio profiles within a first cluster from the candidate audio profiles within a second cluster.

As shown, the audio profile determining engine 114 performs a clustering 222 of the reduced matrix 220 into a plurality of clusters. For example, each column of the matrix 212, corresponding to a vector representation 210 of one of the candidate audio profiles 112 after binning, normalization, and principal component analysis, includes a feature set of features that are represented as rows. The feature space 224 includes a dimensionality that corresponds to the number of features of each vector representation 210, that is, a length of each vector representation 210 and/or a dimension of the matrix 212. The features of each binned, normalized, and PCA-reduced vector representation 210 correspond to a location of the vector representation 210 within the feature space 224. Based on the locations of the vector representations 210, the audio profile determining engine 114 determines a plurality of clusters 226 of vector representations 210. Each cluster 226 includes a number of vector representations 210 that are within a certain proximity to one another within the feature space 224. For example, a first cluster 226-1 includes three of the vector representations 210-1, 210-3, 210-4 that are within a proximity of one another within the feature space 224, and a second cluster 226-2 includes three other vector representations 210-2, 210-5, 210-6 that are also within a proximity of one another within the feature space 224. In various embodiments, the audio profile determining engine 114 performs the clustering 222 according to various clustering techniques, such as a k-medoids clustering technique and/or a Gaussian mixture modeling. In various embodiments, the audio profile determining engine 114 performs the clustering 222 based on a predefined number of clusters 226 (e.g., two clusters). In other various embodiments, the audio profile determining engine 114 also determines a number of clusters 226 by which the vector representations 210 are clustered into a plurality of clusters. For example, the audio profile determining engine 114 can perform a first clustering based on a first number of clusters 226. If the vector representations 210 within each cluster 226 are not within a certain range of tolerance, the audio profile determining engine 114 can perform a second clustering based on a larger number of clusters 226.

As shown, the audio profile determining engine 114 performs a candidate audio profile determination 230 to determine based on the clustering 222 of vector representations 210 within the feature space 224. In various embodiments, for each cluster 226, the audio profile determining engine 114 determines a medoid vector 228, that is, a vector representation 210 of the cluster 226 having a minimal dissimilarity to the other vector representations 210 within the cluster 226. The medoid vector 228 of a cluster 226 represents the candidate audio profile 112 that is the most representative of the candidate audio profiles 112 associated with the cluster 226. For example, for each cluster 226, the audio profile determining engine 114 can determine, for each first vector representation 210 within the cluster 226, an average distance between the first vector representation 210 and each other vector representation 210 associated with the cluster 226. The audio profile determining engine 114 can then determine the medoid vector 228 for each cluster 226 as the first vector representation 210 having the lowest average distance among the calculated average distances of the vector representations 210 of the cluster 226. As shown, the audio profile determining engine 114 determines a first vector representation 210-1 as the medoid vector 228-1 of the first cluster 226-1 and determines a second vector representation 210-2 as the medoid vector 228-2 of the second cluster 226-2.

As shown, the audio profile determining engine 114 determines, by the candidate audio profile determination 230, a number of candidate audio profiles 112 for further evaluation. In various embodiments, the determined candidate audio profiles 112 include the first candidate audio profile 112-1, based on the determination of the first vector representation 210-1 as the first medoid vector 228-1 of the first cluster 226-1, and the second candidate audio profile 112-2, based on the determination of the second vector representation 210-2 as the medoid vector 228-2 of the second cluster 226-2. The audio profile determining engine 114 further evaluates the first candidate audio profile 112-1 and the third candidate audio profile 112-2 in order to determine the audio profile 116 to use for the audio output device 110. The further evaluation is discussed in detail below.

In various embodiments, the audio profile determining engine 114 evaluates the first candidate audio profile 112-1 of the determined plurality of candidate audio profiles 112 through a selection process involving the user. For example, in various embodiments, the device 100 presents a game-style environment to a user and evaluates the candidate audio profiles based on responses of the user. For example, the evaluation can present to the user 120 a multidimensional space 312, such as a virtual reality environment and/or augmented reality environment. Within the multidimensional space 312, the device 100 can display visual indicators 304 (e.g., on a display 302, such as a headset, monitor, or the like) at various intended locations 130, and in which various audio test patterns 310 can be generated by audio output device 110 (e.g., a left speaker 132-1 and a right speaker 132-2) to be perceived at the corresponding intended locations 130. The audio profile determining engine 114 can then ask the user 120 to indicate whether each audio test pattern 310 appears to originate from the same location as the visual indicator 304 within the multidimensional space 312. Based on the responses of the user 120, the audio profile determining engine 114 can determine the clarity and effectiveness of spatial audio generated by the audio output device 110 using the first candidate audio profile 112-1, as perceived by the user 120. An example of the candidate audio profile evaluation process is discussed in detail below in relation to FIGS. 3A-3B and 4A-4B.

FIGS. 3A-3B are an illustration of a first step of an audio profile selection by the device of FIG. 1, according to various embodiments. In various embodiments, the first step of the audio profile selection is performed by the audio profile determining engine 114 of FIG. 1. In various embodiments, the audio profile selection is based on the determination of candidate audio profiles 112 as shown in FIG. 2.

As shown, in FIG. 3A, at a first time, the audio profile determining engine 114 generates an audio test pattern 310 that is intended to be perceived by a user 120 as occurring at a first intended location 130-1. In various embodiments, the audio profile determining engine 114 applies the first candidate audio profile 112-1 to an audio input 126 to cause the left speaker 132-1 to generate a left audio output 128-1, and to cause the right speaker 132-2 to generate a right audio output 128-2. The combination of the left audio output 128-1 and the right audio output 128-2, based on the first candidate audio profile 112-1, generates the audio test pattern 310 that the user 120 should perceive at the first intended location 130-1. Concurrently, the audio profile determining engine 114 displays a visual indicator 304 on the display 302 that corresponds to the first intended location 130-1. The audio profile determining engine 114 presents, to the user 120, a first inquiry 306-1 as to whether the sound is originating from the same location as the visual indicator 304 (e.g., the first intended location 130-1). The audio profile determining engine 114 receives, from the user 120, a first response 308-1 including a user agreement, confirming that the user 120 perceives the sound as originating from the same location as the visual indicator 304. Based on the first response 308-1, the audio profile determining engine 114 continues evaluating the first candidate audio profile 112-1.

As shown, in FIG. 3B, at a second time, the audio profile determining engine 114 generates an audio test pattern 310 that is intended to be perceived by the user 120 as occurring at a second intended location 130-2. In various embodiments, the audio profile determining engine 114 applies the first candidate audio profile 112-1 to an audio input 126 to cause the left speaker 132-1 to generate a left audio output 128-1, and to cause the right speaker 132-2 to generate a right audio output 128-2. The combination of the left audio output 128-1 and the right audio output 128-2, based on the first candidate audio profile 112-1, generates the audio test pattern 310 that the user 120 should perceive at the second intended location 130-2. Concurrently, the audio profile determining engine 114 displays a visual indicator 304 on the display 302 that corresponds to the second intended location 130-2. The audio profile determining engine 114 presents, to the user 120, a second inquiry 306-2 as to whether the sound is originating from the same location as the visual indicator 304 (e.g., the second intended location 130-2). The audio profile determining engine 114 receives, from the user 120, a second response 308-2 including a user disagreement, indicating that the user 120 does not perceive the sound as originating from the same location as the visual indicator 304. Based on the second response 308-2, the audio profile determining engine 114 determines that the first candidate audio profile 112-1 is not to be used as the audio profile 116 for the audio output device 110. Instead, the audio profile determining engine 114 proceeds with a second step of the audio profile selection in which another candidate audio profile 112 is evaluated.

FIGS. 4A-4B are an illustration of a second step of an audio profile selection by the device of FIG. 1, according to various embodiments. In various embodiments, the second step of the audio profile selection is performed by the audio profile determining engine 114 of FIG. 1. In various embodiments, the audio profile selection is based on the determination of candidate audio profiles 112 as shown in FIG. 2.

As shown, in FIG. 4A, at a third time, the audio profile determining engine 114 generates an audio test pattern 310 that is intended to be perceived by the user 120 as occurring at the first intended location 130-1. In various embodiments, the audio profile determining engine 114 applies the second candidate audio profile 112-2 to the audio input 126 to cause the left speaker 132-1 to generate a left audio output 128-1, and to cause the right speaker 132-2 to generate a right audio output 128-2. The combination of the left audio output 128-1 and the right audio output 128-2, based on the second candidate audio profile 112-2, generates the audio test pattern 310 that the user 120 should perceive at the first intended location 130-1. Concurrently, the audio profile determining engine 114 displays a visual indicator 304 on the display 302 that corresponds to the first intended location 130-1. The audio profile determining engine 114 presents, to the user 120, a third inquiry 306-3 as to whether the sound is originating from the same location as the visual indicator 304 (e.g., the first intended location 130-1). The audio profile determining engine 114 receives, from the user 120, a third response 308-3 including a user agreement, confirming that the user 120 perceives the sound as originating from the same location as the visual indicator 304. Based on the third response 308-3, the audio profile determining engine 114 continues evaluating the second candidate audio profile 112-2.

As shown, in FIG. 4B, at a fourth time, the audio profile determining engine 114 generates an audio test pattern 310 that is intended to be perceived by the user 120 as occurring at the second intended location 130-2. In various embodiments, the audio profile determining engine 114 applies the second candidate audio profile 112-2 to an audio input 126 to cause the left speaker 132-1 to generate a left audio output 128-1, and to cause the right speaker 132-2 to generate a right audio output 128-2. The combination of the left audio output 128-1 and the right audio output 128-2, based on the second candidate audio profile 112-2, generates the audio test pattern 310 that the user 120 should perceive at the second intended location 130-2. Concurrently, the audio profile determining engine 114 displays a visual indicator 304 on the display 302 that corresponds to the second intended location 130-2. The audio profile determining engine 114 presents, to the user 120, a fourth inquiry 306-4 as to whether the sound is originating from the same location as the visual indicator 304 (e.g., the second intended location 130-2). The audio profile determining engine 114 receives, from the user 120, a fourth response 308-4 including a user agreement, confirming that the user 120 perceives the sound as originating from the same location as the visual indicator 304. Based on the fourth response 308-4, the audio profile determining engine 114 determines that the second candidate audio profile 112-2 is to be used as the audio profile 116 for the audio output device 110.

In various embodiments, the audio profile determining engine 114 can perform the candidate audio profile evaluation process, such as shown in FIGS. 3A-3B and 4A-4B, in various ways. For example, in various embodiments, the audio profile determining engine 114 can present the visual indicators 304 in various ways, such as a symbol shown within the multidimensional space 312, or a character or object that is the source of the sound comprising the audio test pattern 310. In various embodiments, rather than generating visual indicators 304, the audio profile determining engine 114 could generate each inquiry 306 as a question about the perceived location of the audio test pattern 310 (e.g.: “Does the sound seem to be near your left ear?”) In various embodiments, rather than generating inquiries 306, the audio profile determining engine 114 could generate audio test pattern 310 and determine the response 308 of the user 120 based on user input received from the user 120. For example, the audio profile determining engine 114 can ask the user 120 to move his or her head 122 to look at the location at which the audio test pattern 310 is perceived to be originating. Based on sensor feedback (e.g., a head-tracking camera that visually determines a head orientation of the 120, or an eye-tracking camera that visually determines the eye-gaze orientation of the 120, or an orientation sensor included in a helmet worn by the user 120), the audio profile determining engine 114 can determine whether the user 120 is looking toward the intended location 130 or is looking elsewhere. As another example, the audio profile determining engine 114 can ask the user 120 to point toward the location at which the user 120 perceives the audio test pattern 310 to be originating. Based on sensor feedback (e.g., a hand-tracking camera that visually determines a hand orientation of the 120, or an orientation sensor included in a glove worn by the user 120), the audio profile determining engine 114 can determine whether the user 120 is pointing toward the intended location 130 or is pointing elsewhere.

In various embodiments, the audio profile determining engine 114 can perform the candidate audio profile evaluation process of various candidate audio profiles 112 in various ways. As shown in FIGS. 3A-3B, the audio profile determining engine 114 can perform a first step including evaluating a first candidate audio profile 112-1. Based on the responses 308 of the user 120 during the first step, the audio profile determining engine 114 can either determine the first candidate audio profile 112-1 as the audio profile 116 for the audio output device 110, or discard the first candidate audio profile 112-1 and continue to the second step to evaluate a second candidate audio profile 112-2. As another example, in various embodiments, the audio profile determining engine 114 can evaluate each of at least two candidate audio profiles 112 and then determine the audio profile 116 based on a comparison of the responses 308 of the user 120 to each of the at least two candidate audio profiles 112. For example, the audio profile determining engine 114 can assign a score to each of two or more candidate audio profiles 112 based on the responses 308 of the user 120, and then select the candidate audio profile 112 that has been assigned a higher or highest score. As yet another example, in various embodiments, the audio profile determining engine 114 can concurrently evaluate each of at least two candidate audio profiles 112. For example, the audio profile determining engine 114 can generate a first audio test pattern 310 based on a first candidate audio profile 112-1 (e.g., a tone at a first time) and a second audio test pattern 310 based on a second candidate audio profile 112-2 (e.g., a tone at a second time) and then present to the user 120 an inquiry 306 that asks which audio test pattern 310 more closely matches the intended location 130-1 of the visual indicator 304. Based on the response 308 of the user 120 indicating a preference or selection of one of the audio test patterns 310 (e.g., the second tone), the audio profile determining engine 114 can select one of the candidate audio profiles 112 as the audio profile 116 for the audio output device 110. As yet another example, the audio profile determining engine 114 can generate several audio test patterns 310 for the user 120, and then receive, from the user 120, one or more responses 308 that indicate a user preference ranking of the audio test patterns 310. Based on the responses 308, the audio profile determining engine 114 can determine a user preference ranking of the at least two candidate audio profiles 112, and can determine the audio profile 116 for the audio output device 110 based on the user preference ranking of the at least two candidate audio profiles 112. As yet another example, the audio profile determining engine 114 can determine, among the at least two candidate audio profiles 112, the candidate audio profile 112 for which the locations indicated by the user 120 more closely or most closely match the intended locations 130 of the corresponding audio test patterns 310.

In some cases, the responses 308 of the user 120 could indicate that neither or none of two or more audio test patterns 310 matches the locations of the visual indicators 304. For example, the user input received from the user 120 could indicate that the user 120 does not perceive the audio as originating from an intended location, that the user 120 perceives the audio as originating from a location other than the intended location, or that scores received from the user 120 are not above a threshold. Based on the responses 308 of the user 120, the device 100 could determine that neither or none of two or more candidate audio profiles 112 used to present the audio test patterns 310 to the user 120 causes the audio output device 110 to generate clear and effective spatial audio for the user 120. In various embodiments, the audio profile determining engine 114 can determine that the responses 308 of the user 120 indicate a rejection of the two or more candidate audio profiles 112 that were determined based on the plurality of clusters 226. Based on the rejection, the audio profile determining engine 114 can re-cluster the vector representations 210, excluding the two or more vector representations 210 that correspond to the candidate audio profile 112 that were determined for evaluation based on the first plurality of clusters 226. Based on the re-clustering, the audio profile determining engine 114 can determine two or more updated clusters 226. The audio profile determining engine 114 can determine another vector representation 210 for each of the two or more updated clusters 226 (e.g., a medoid vector 228 of each of the two or more updated clusters 226). The audio profile determining engine 114 can perform another round of evaluation based on the candidate audio profiles 112 corresponding to the two or more another vector representations 210.

FIG. 5 illustrates a flow diagram of method steps for determining an audio profile for an audio output device, according to various embodiments. In various embodiments, at least some of the method steps of FIG. 5 are performed by the audio profile determining engine 114 and/or the audio rendering engine 118 of FIG. 1. Although the method steps are described with respect to the systems of FIGS. 1 through 4B, persons skilled in the art will understand that any system configured to perform the method steps, in any order, falls within the scope of the various embodiments.

As shown, a method 500 begins at step 502 in which the audio profile determining engine generates a vector representation of each candidate audio profile of a plurality of candidate audio profiles. In various embodiments, each vector representation aggregates two or more left ear samples and two or more right ear samples. In various embodiments, each vector representation concatenates an average left ear sample and an average right ear sample. In various embodiments, the vector representations of the candidate audio profiles are further processed, such as by aggregation into a matrix, binning, normalization, and/or a principal component analysis. In various embodiments, generating the vector representations can be performed according to at least some of the method steps of the flow diagram of FIG. 6.

At step 504, the audio profile determining engine clusters the vector representations of the candidate audio profiles into a plurality of clusters. In various embodiments, the audio profile determining engine determines the locations of the vector representations within a feature space and determines the clusters of vectors that are within a proximity of one another. In various embodiments, the audio profile determining engine determines the clusters based on a clustering technique, such as a k-medoids clustering technique. In various embodiments, the audio profile determining engine clusters the vector representations according to a predefined number of clusters (e.g., two clusters). In various embodiments, the clustering can be performed according to at least some of the method steps of the flow diagram of FIG. 6.

At step 506, the audio profile determining engine presents, to a user, two or more audio test patterns, wherein each audio test pattern is based on one or more candidate audio profiles that are associated with a medoid vector of one cluster of the plurality of clusters. In various embodiments, the audio profile determining engine presents the two or more audio test patterns to the user. In various embodiments, the audio profile determining engine generates each audio test pattern to be perceived by the user at an intended location within a multidimensional space (e.g., a virtual reality environment or augmented reality environment), based on one of the candidate audio profiles. In various embodiments, the audio profile determining engine concurrently displays a visual indicator at the intended location within the multidimensional space. Alternatively or additionally, in various embodiments, the audio profile determining engine asks the user to indicate the location within the multidimensional space where the user perceives the audio test pattern to originate.

At step 508, the audio profile determining engine receives, from the user, at least one response based on the two or more audio test patterns. In various embodiments, the audio profile determining engine receives either a user agreement or a user disagreement as to whether the user perceives the audio test pattern to originate from the same location as a displayed visual indicator. In various embodiments, the audio profile determining engine detects a location where the user is looking or pointing, as the location where the user perceives each audio test pattern to originate, and determines whether each location indicated by the user match the intended location of each audio test pattern.

At step 510, the audio profile determining engine determines that the candidate audio profile associated with one of the audio test patterns is to be used as the audio profile for the audio output device. In various embodiments, the audio profile determining engine determines the audio profile as the candidate audio profile for which the locations indicated by the user more closely or most closely match the intended locations of the audio test patterns. In various embodiments, the audio profile determining engine determines the audio profile as the candidate audio profile having a highest user preference ranking among the candidate audio profiles.

At step 512, the audio profile determining engine determines an audio profile for the audio output device based on the at least one response of the user. In various embodiments, the audio profile determining engine determines the audio profile as one of the candidate audio profiles for which the user indicated a user agreement with the presented audio test patterns. In various embodiments, the audio profile determining engine determines a user preference ranking of the at least two candidate audio profiles for which the audio profile determining engine presented audio test patterns.

At step 514, the audio rendering engine causes the audio output device to output audio based on the audio profile. In various embodiments, the audio rendering engine renders spatial audio based on the audio profile, wherein the combination of a left audio output of a left speaker and a right audio output of a right speaker cause the user to perceive an audio output as originating at an intended location relative to the head of the user.

At step 516, the audio profile determining engine excludes the at least two candidate audio profiles from the plurality of candidate audio test patterns. The audio profile determining engine then returns to step 504 to determine another candidate audio profile (e.g., at least two other candidate audio profiles) based on a re-clustering of the plurality of candidate audio profiles, excluding the first at least two candidate audio profiles.

FIG. 6 illustrates a flow diagram of method steps for determining one or more candidate audio profiles for an audio output device, according to various embodiments. In various embodiments, at least some of the method steps of FIG. 6 are performed by the audio profile determining engine 114 of FIG. 1. In various embodiments, the method steps of the flow diagram of FIG. 6 can be performed at steps 502 and 504 of FIG. 5. Although the method steps are described with respect to the systems of FIGS. 1 through 5, persons skilled in the art will understand that any system configured to perform the method steps, in any order, falls within the scope of the various embodiments.

As shown, a method 600 begins at step 602 in which the audio profile determining engine determines an average of two or more left ear samples and an average of two or more right ear samples of each candidate audio profile. In various embodiments, the averaging can involve a determination of a mathematical mean or median of the two or more left ear samples to determine the average of the two or more left ear samples, and a determination of a mathematical mean or median of the two or more right ear samples to determine the average of the two or more right ear samples (e.g., the impulse response and/or frequency response of the left ear of a user to all audio cues and/or audio frequencies). The average of the left ear samples can represent an average HRIR and/or an average HRTF of the left ear samples of the candidate audio profile. The average of the right ear samples can represent an average HRIR and/or an average HRTF of the right ear samples of the candidate audio profile (e.g., the impulse response and/or frequency response of the right ear of a user to all audio cues and/or audio frequencies).

At step 604, the audio profile determining engine combines the average of the two or more left ear samples and the average of the two or more right ear samples of each candidate audio profile to form a vector representation. In various embodiments, the combining can include concatenating the average of the two or more left ear samples and the average of the two or more right ear samples.

At step 606, the audio profile determining engine generates a matrix including the vector representation of each candidate audio profile. In various embodiments, the generating includes combining a one-dimensional vector representation of each candidate audio profile along a second dimension of the matrix.

At step 608, the audio profile determining engine performs binning of the matrix. In various embodiments, the audio profile determining engine generates one or more bins, each representing a frequency range within a frequency spectrum of the matrix. In various embodiments, the audio profile determining engine generates one or more bins, each representing a frequency range within a frequency spectrum of the matrix. In various embodiments, the bins can cover only a portion of the audible frequency spectrum (e.g., 1 kilohertz to 14 kilohertz), and other frequencies that are above or below the portion of the audible frequency spectrum can be discarded.

At step 610, the audio profile determining engine performs a normalization of the matrix. In various embodiments, for each vector element of each vector representation or column of the matrix, the audio profile determining engine calculates a logarithmic value of the vector element, such as a normalized logarithmic intensity of a frequency response for each frequency bin within a binned human-audible frequency range. In various embodiments, the audio profile determining engine normalizes the matrix in other ways, such as adding a positive or negative offset or bias to the vector element and/or clipping the vector element based on a high or low clipping value.

At step 612, the audio profile determining engine performs a principal component analysis of the matrix. In various embodiments, the audio profile determining engine determines, among a feature set of the binned and normalized matrix, a reduced feature set of features that are representative of the matrix. In various embodiments, the audio profile determining engine determines, among the feature set of the binned and normalized matrix, an excludable feature set of features that are not representative of the matrix. In various embodiments, the audio profile determining engine retains a reduced feature set and exclude the excluded feature set of the binned and normalized matrix to generate a reduced matrix.

At step 614, the audio profile determining engine positions each vector representation of the matrix in a feature space. In various embodiments, the feature space includes a dimensionality that corresponds to the number of features of each vector representation, that is, a length of each vector representation.

At step 616, the audio profile determining engine determines one or more clusters of vector representations that are close to one another in the feature space. In various embodiments, the clustering groups the vectors based on their distance to other vectors within the feature space and identifies each cluster based on the vectors that are within a certain distance of other vectors in the feature space. In various embodiments, the clustering includes one or more clustering techniques, such as k-medoids clustering technique and/or a Gaussian mixture modeling technique.

At step 618, the audio profile determining engine determines, for each cluster of the one or more clusters, a medoid vector among the vector representations of the cluster. In various embodiments, the medoid vector is the vector representation of the cluster having a minimal dissimilarity to the other vector representations within the cluster. In various embodiments, the medoid vector of a cluster represents the candidate audio profile that is the most representative of the candidate audio profiles associated with the cluster.

At step 620, the audio profile determining engine determines, for further evaluation, the candidate audio profile associated with the medoid vector of each cluster of the one or more clusters. In various embodiments, the determined candidate audio profiles are further evaluated by a selection process involving the user. In various embodiments, the selection process includes method steps 506-516 of FIG. 5.

In sum, techniques for selecting an audio profile for a user include generating a vector representation of each candidate audio profile of a plurality of candidate audio profiles and clustering the vector representations into a plurality of clusters. Clustering the vector representations enables a determination of which candidate audio profiles are highly representative among the candidate audio profiles associated with each cluster. The techniques also include determining an audio profile for the user based on the plurality of clusters. Determining the audio profile based on the plurality of clusters enables a determination of the audio profile that is likely to cause the spatial audio generated by the device to be accurately perceived by the user. The techniques also include presenting, to the user, audio test patterns that are each based on one or more candidate audio profiles that are associated with one of the clusters. Based on responses received from the user to the audio test patterns, an audio profile is determined and used to present audio to the user. Selecting the audio profile based on user responses to the presented audio test patterns can allow the audio output device to be configured with a suitable audio profile through a simplified and enjoyable user experience.

At least one technical advantage of the disclosed techniques relative to the prior art is that, with the disclosed techniques, a user can be quickly and effectively guided through the process of selecting an effective audio profile usable by an audio output device to generate spatial audio for the user. The disclosed techniques further increase the likelihood that the user will select an effective audio profile so that an audio output device is able to generate improved spatial audio that spatial audio using audio profiles selected by other techniques. The disclosed techniques also reduce the computing resources needed to select candidate audio profiles from a potentially large number of audio profiles while also improving the likelihood that a candidate profiles will be effective for the user. The ability to select better candidate profiles reduces the number of candidate profiles that have to be considered during the audio profile selection process, which further reduces the time spent selecting an audio profile and the computing resources used to select the audio profile. These technical advantages provide one or more technological improvements over prior art approaches.