WO2019027812A1 - Classification d'objet audio sur la base de métadonnées de localisation - Google Patents

Classification d'objet audio sur la base de métadonnées de localisation Download PDF

Info

Publication number
WO2019027812A1
WO2019027812A1 PCT/US2018/043980 US2018043980W WO2019027812A1 WO 2019027812 A1 WO2019027812 A1 WO 2019027812A1 US 2018043980 W US2018043980 W US 2018043980W WO 2019027812 A1 WO2019027812 A1 WO 2019027812A1
Authority
WO
WIPO (PCT)
Prior art keywords
audio
audio object
dialog
location metadata
objects
Prior art date
Application number
PCT/US2018/043980
Other languages
English (en)
Inventor
Mark William GERRARD
Original Assignee
Dolby Laboratories Licensing Corporation
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Dolby Laboratories Licensing Corporation filed Critical Dolby Laboratories Licensing Corporation
Priority to CN201880049177.7A priority Critical patent/CN110998724B/zh
Priority to EP18747091.9A priority patent/EP3662470B1/fr
Priority to US16/636,241 priority patent/US11386913B2/en
Publication of WO2019027812A1 publication Critical patent/WO2019027812A1/fr

Links

Classifications

    • G—PHYSICS
    • G10—MUSICAL INSTRUMENTS; ACOUSTICS
    • G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
    • G10L25/00—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00
    • G10L25/48—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 specially adapted for particular use
    • G10L25/51—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 specially adapted for particular use for comparison or discrimination
    • G—PHYSICS
    • G10—MUSICAL INSTRUMENTS; ACOUSTICS
    • G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
    • G10L21/00—Speech or voice signal processing techniques to produce another audible or non-audible signal, e.g. visual or tactile, in order to modify its quality or its intelligibility
    • G10L21/02—Speech enhancement, e.g. noise reduction or echo cancellation
    • G—PHYSICS
    • G10—MUSICAL INSTRUMENTS; ACOUSTICS
    • G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
    • G10L25/00—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00
    • G10L25/03—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 characterised by the type of extracted parameters
    • G—PHYSICS
    • G10—MUSICAL INSTRUMENTS; ACOUSTICS
    • G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
    • G10L25/00—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00
    • G10L25/78—Detection of presence or absence of voice signals

Definitions

  • the present disclosure generally relates to the field of audio processing, and more specifically to methods, apparatuses and systems that classify audio objects based upon location metadata.
  • Audio formats may be employed to reproduce an audio scene.
  • the different audio channels may for example be intended to be played back by respective loudspeakers for together reproducing the audio scene.
  • Audio channels are typically associated with a certain position in space, or with a certain loudspeaker (such as a left, right or center speaker).
  • audio formats have been introduced in which the traditional audio channels are supplemented (or even replaced) by audio objects which are associated with time- variable spatial positions.
  • An audio object typically comprises audio content and associated location metadata. While the audio content represents the actual audio signal (or the actual audio stream), the location metadata defines a position of the audio object in an audio scene.
  • Many different audio processing systems have been proposed for coding, decoding, or rendering of audio represented via such audio formats.
  • a desirable property for such audio formats and audio processing systems is to enable faithful reconstruction of an original audio scene. It is therefore desirable to avoid or mitigate coding artefacts or other artefacts caused by audio processing systems and/or audio reproduction systems.
  • Other desirable properties may for example include bitrate efficiency and/or computational efficiency. However, it is often difficult to provide high bitrate efficiency and/or high computational efficiency while still providing a faithful reconstruction without audible artefacts.
  • One approach for making a suitable tradeoff between these sometimes conflicting properties is to treat different types of audio signals differently.
  • the document US 2016/0078879 A1 (which is incorporated herein by reference in its entirety) discloses apparatuses and methods for classifying and processing audio signals. An audio classifier classifies an audio signal into at least one audio type in real time.
  • An audio improving device processes the audio signal for improving a listening experience of an audience.
  • An adjusting unit adjusts at least one parameter of the audio improving device in a continuous manner based on a confidence value of the at least one audio type.
  • the audio improving unit employs dialog enhancement when speech is present in the audio signal, and the dialog enhancement is turned off when speech is absent in the audio signal.
  • the document US 2015/0332680 A1 (which is incorporated herein by reference in its entirety) discloses a method in which perceptual importance of audio objects is determined, and in which the audio objects are clustered based on their perceptual importance.
  • This document also discloses an audio classification component which employs a set of pre-trained models representing statistical properties of respective target audio types (such as dialog, music, sound effects and noise) to compute confidence scores for each target audio type, and estimates the best matched audio types.
  • the confidence scores are computed using different machine learning methods.
  • the document US 2010/0121634 A1 (which is incorporated herein by reference in its entirety) discloses methods and apparatus for speech enhancement in entertainment audio.
  • the entertainment audio is analyzed to classify time segments of the audio as being either speech or other audio.
  • Speech enhancement is applied to time segments classified as speech.
  • Figure 1 is a perspective view of a room with an example loudspeaker setup for reproducing an audio scene
  • Figure 2 shows an encoder side system for classifying audio objects based on location metadata, according to an example embodiment
  • Figure 3 shows a decoder side system for classifying audio objects based on location metadata, according to an example embodiment
  • Figure 4 shows a system for classifying audio objects based on both location metadata and audio content, according to an example embodiment
  • Figure 5 shows a system for classifying audio objects and for clustering audio objects based on the audio object classification, according to an example embodiment
  • Figure 6 shows a system for classifying audio objects and for enhancing audio signals based on the audio object classification, according to an example embodiment
  • Figure 7 is a flow chart of a method for classifying audio objects based on location metadata, according to an example embodiment
  • Figure 8 is a flow chart of a method for classifying audio objects and for performing dialog enhancement based on the audio object classification, according to an example embodiment
  • Figure 9 is a flow chart of a method for classifying audio objects and for performing clustering based on the audio object classification, according to an example embodiment
  • Figure 10 is a top view of the room in figure 1 ;
  • Figure 1 1 is a side view of the room in figure 1 .
  • an audio signal may be a pure audio signal, an audio part of an audiovisual signal or multimedia signal, or any of these in combination with metadata.
  • a channel is an audio signal associated with a predefined/fixed spatial position/orientation or an undefined spatial position such as "left” or "right”.
  • an audio object or audio object signal is an audio signal associated with a spatial position susceptible of being time-variable, in other words a spatial position whose value may be re-assigned or updated over time.
  • example embodiments provide a method.
  • the method comprises receiving location metadata associated with an audio object.
  • the location metadata defines a position of the audio object in an audio scene.
  • the method further comprises estimating, based on the location metadata, whether the audio object includes dialog, and assigning a value to an object type parameter representative of a result of the estimation.
  • an audio object which includes dialog may for example include speech or voice content.
  • the object type parameter may indicate a level of confidence that the audio object includes dialog.
  • the object type parameter may for example be a number (for example a float number).
  • a high value of the object type parameter may for example indicate a high confidence that the audio object includes dialog, and a low value of the object type parameter may for example indicate that the audio object is less likely to include dialog.
  • a low value of the object type parameter may for example indicate a high confidence that the audio object includes dialog, and a high value of the object type parameter may for example indicate that the audio object is less likely to include dialog.
  • the object type parameter may be a Boolean type parameter indicating whether or not a level of confidence that the audio object includes dialog is above (or exceeds) a threshold.
  • the object type parameter being zero may for example indicate that the confidence level is below the threshold, and the object type parameter being one may for example indicate that the confidence level is above (or exceeds) the threshold.
  • the object type parameter being one may for example indicate that the confidence level is below the threshold, and the object type parameter being zero may for example indicate that the confidence level is above (or exceeds) the threshold.
  • the estimation may be performed based on a position of the audio object in a front-back direction of the audio scene.
  • the position in the front-back direction may be defined by the location metadata.
  • Audio objects including dialog are often located at the front of the audio scene.
  • An audio object located at the back of the audio scene may be less likely to include dialog.
  • the front of the audio scene may be in front of an intended listener, and the back of the audio scene may be behind the intended listener.
  • the front of the audio scene may for example be a position associated with a screen (such as a movie/cinema screen or TV screen) at which images (for example in the form of a video stream) associated with the audio scene are supposed to be displayed.
  • Audio objects including dialog may for example be located at the screen, or at the edge of the screen.
  • estimating whether the audio object includes dialog may comprise associating a position at a front of the audio scene with a higher level of confidence that the audio object includes dialog than levels of confidence associated with positions further back in the audio scene.
  • an audio object with a position at the front of the audio scene may be assigned (or attributed) a higher confidence level that it includes dialog than confidence levels assigned (or attributed) to audio objects located further back in the audio scene.
  • the confidence level that the audio object incudes dialog may decrease. It will be appreciated that other factors may also influence the estimation whether an audio object includes dialog, and that the confidence level may therefore decrease even if an audio object moves towards the front of the audio scene.
  • estimating whether the audio object includes dialog may comprise computing a speed (or velocity) of the audio object based on location metadata associated with different time frames, and estimating, based on the speed (or velocity), whether the audio object includes dialog.
  • Audio objects including dialog often do not move around that much in the audio scene. Objects only including dialog may for example be stationary in the audio scene. An audio object moving rapidly (or with high speed) in the audio scene may be less likely to include dialog.
  • estimating whether the audio object includes dialog may comprise associating a first value of the speed with a higher level of confidence that the audio object includes dialog than a level of confidence associated with a second value of the speed.
  • the first value of the speed may be lower than the second value of the speed.
  • an audio object with a low (or zero) speed may be assigned (or attributed) a higher confidence level that it includes dialog than confidence levels assigned (or attributed) to audio objects with higher speed.
  • the confidence level that the audio object incudes dialog may decrease. It will be appreciated that other factors may also influence the estimation whether an audio object includes dialog, and that the confidence level may therefore decrease even if the speed of the audio object decreases.
  • estimating whether the audio object includes dialog may comprise computing an acceleration of the audio object based on location metadata associated with different time frames, and estimating, based on the acceleration, whether the audio object includes dialog.
  • An audio objects with high/large acceleration may be less likely to include dialog than audio objects with low/small (or zero) acceleration.
  • Estimating whether the audio object includes dialog may for example comprise associating a first value of the acceleration with a higher level of confidence that the audio object includes dialog than a level of confidence associated with a second value of the acceleration.
  • the first value of the acceleration may for example be lower than the second value of the acceleration.
  • an audio object with a low (or zero) acceleration may for example be assigned (or attributed) a higher confidence level that it includes dialog than confidence levels assigned (or attributed) to audio objects with higher acceleration.
  • the confidence level that the audio object incudes dialog may for example decrease. It will be appreciated that other factors may also influence the estimation whether an audio object includes dialog, and that the confidence level may therefore decrease even if the acceleration of the audio object decreases.
  • the estimation may be performed based on a level of elevation of the audio object defined by the location metadata.
  • Audio objects including dialog are often located at an intended listener level, or at a floor level.
  • An audio object located at an elevated position above the intended listener in the audio scene is often less likely to include dialog.
  • estimating whether the audio object includes dialog may comprise associating a first level of elevation of the audio object with a higher level of confidence that the audio object includes dialog than levels of confidence associated with other levels of elevation of the audio object.
  • the first level of elevation may correspond to a floor level of the audio scene or a vertical position of an intended listener.
  • an audio object having a similar vertical position as an intended listener (or being located at a floor level) may be assigned (or attributed) a higher confidence level that it includes dialog than confidence levels assigned (or attributed) to audio objects located higher up or down in the audio scene.
  • the confidence level that the audio object incudes dialog may for example decrease. It will be appreciated that other factors may also influence the estimation whether an audio object includes dialog, and that the confidence level may therefore decrease even the audio object moves closer to the intended vertical position of the listener.
  • estimating whether the audio object includes dialog may comprise computing a linear combination (or a weighted sum) of at least a position of the audio object, and a speed of the audio object.
  • the method may comprise setting initial coefficients for respective terms in the linear combination (or weighted sum), estimating, using the linear combination, whether a plurality of audio objects include dialog, and adjusting coefficients for respective terms in the linear combination based on the estimations and based on knowledge of whether the respective audio objects actually include dialog. In this way, the coefficients in the linear combination may be tuned for improving the reliability (or accuracy) of the estimation.
  • the method may comprise receiving a plurality of audio objects.
  • Each of the received audio objects may include audio content and location metadata.
  • the location metadata of an audio object (for example each of the received audio objects) may define a position of that audio object in an audio scene.
  • the method may comprise estimating, based on the location metadata of the respective audio objects, whether the respective audio objects include dialog, assigning values to object type parameters representative of results of the respective estimations, and selecting a subset of the plurality of audio objects based on the assigned values of the object type parameters.
  • the subset may include one or more audio objects.
  • the selected subset may for example include those audio objects which are most likely to include dialog.
  • the method may comprise subjecting at least one audio object in the selected subset to dialog enhancement.
  • Dialog enhancement may for example be applied to clarify or increase intelligibility of a dialog (or of speech) for making the dialog easier to hear and understand.
  • dialog or of speech
  • ways to enhance dialog or speech
  • Examples of ways to enhance dialog may for example be found in paragraph [0208] of US 2016/0078879 A1 (which is incorporated herein by reference in its entirety), and in paragraphs [0027]-[0028] in US 2010/0121634 A1 (which is incorporated herein by reference in its entirety).
  • the method may comprise performing clustering such that the audio content from those of the plurality of audio objects outside the selected subset is included in a collection of clusters and such that at least one audio object of the selected subset is excluded from the clustering or the audio content of at least one audio object of the selected subset is included in a cluster which does not include audio content from any of those of the plurality of audio objects outside the selected subset (or which only includes audio content from audio objects in the selected subset).
  • While clustering may be an efficient way to reduce the complexity of an audio scene with many audio objects (for example to improve bitrate efficiency for transmission between an encoder side and a decoder side), dialog may be perceived as so important that it should be kept separate from other audio objects. Therefore, it may be desirable to exclude dialog audio objects from the clustering, or to cluster dialog audio objects separately from the audio objects not including dialog.
  • the clustering may for example be based on the location metadata. Audio objects which are located close together in the audio scene may for example be combined into the same cluster during the clustering.
  • the method may comprise, for each of the one or more audio objects in the selected subset, analyzing the audio content of the audio object and determining, based on the analysis, a value indicating a level of confidence that the audio object includes dialog.
  • analysis of the audio content of an audio object may provide a more reliable estimation of whether the audio object includes dialog.
  • the confidence level obtained via analysis of the audio content of an audio object may therefore be referred to as a refined confidence level. Since analysis of the audio content of a large number of audio objects may have high computational complexity, it may be advantageous to employ the object type parameters (which are obtained based on the location metadata of the audio objects) to reduce the number of audio objects to be subjected to such audio content analysis.
  • the analysis of the audio content may for example include spectral flux detection, and/or use of a machine learning algorithm.
  • the documents US 2016/0078879 A1 (which is incorporated herein by reference in its entirety, see in particular paragraphs [0405], [0431 ] and [0443] therein), US 2015/0332680 A1 (which is incorporated herein by reference in its entirety) and US 2010/01 21634 A1 (which is incorporated herein by reference in its entirety, see in particular paragraph [0021 ] therein) referred to in the background section also provide examples of ways to analyze audio content to estimate whether an audio signal includes dialog (or speech).
  • the method may comprise subjecting at least one audio object from the selected subset to dialog enhancement.
  • the degree of dialog enhancement to which the at least one audio object is subjected may be determined based on the corresponding at least one determined value.
  • dialog enhancement may for example be applied to that audio object to make the dialog in that audio object easier to hear and/or distinguish.
  • the degree of dialog enhancement may for example be time-dependent and/or frequency-dependent.
  • the selected subset may include multiple audio objects.
  • the method may comprise selecting at least one audio object from the selected subset based on the determined values, and subjecting the selected at least one audio object to dialog enhancement.
  • the object type parameter may for example indicate that an audio object includes dialog, but a closer analysis of the audio content of the audio object may reveal that the audio object in fact does not include dialog. Dialog enhancement may therefore be omitted for such an audio object.
  • the object type parameter may for example indicate that an audio object includes dialog, but a closer analysis of the audio content of the audio object may reveal that the audio object in fact does not include dialog. Such an audio object may therefore be included in the clustering just like the other audio objects not including dialog, while the audio objects actually including dialog may be excluded from the clustering or may be clustered separately.
  • the method may comprise outputting a bitstream.
  • the bitstream may include the audio content, the location metadata, and the assigned valued of the object type parameter (which is determined based on the location metadata of the audio object).
  • the method may comprise outputting a bitstream.
  • the bitstream may comprise the audio content, the location metadata, and the determined value (which is determined based on analysis of the audio content of the audio object) indicating a level of confidence that the audio object includes dialog.
  • example embodiments provide a system configured to receive location metadata associated with an audio object.
  • the location metadata defines a position of the audio object in an audio scene.
  • the system comprises a processing section configured to estimate, based on the location metadata, whether the audio object includes dialog, and to assign a value to an object type parameter representative of a result of the estimation.
  • the system may for example be configured to perform the method of any of the example embodiments of the first aspect.
  • the system may be configured to receive a plurality of audio objects.
  • Each of the audio objects may include audio content and location metadata.
  • the location metadata of the audio objects may define positions of the respective audio objects in an audio scene.
  • the processing section may be configured to estimate, based on the respective location metadata, whether the respective audio objects include dialog, and assign values to object type parameters representative of results of the respective estimations.
  • the system may further comprise a selection section configured to select a subset of the plurality of audio objects based on the assigned values of the object type parameters. The subset may include one or more audio objects.
  • the system may comprise a dialog enhancement section configured to subject at least one audio object in the selected subset to dialog enhancement.
  • the system may comprise a clustering section configured to perform clustering such that the audio content from those of the plurality of audio objects outside the selected subset is included in a collection of clusters and such that at least one audio object of the selected subset is excluded from the clustering or the audio content of at least one audio object of the selected subset is included in a cluster which does not include audio content from any of those of the plurality of audio objects outside the selected subset.
  • a clustering section configured to perform clustering such that the audio content from those of the plurality of audio objects outside the selected subset is included in a collection of clusters and such that at least one audio object of the selected subset is excluded from the clustering or the audio content of at least one audio object of the selected subset is included in a cluster which does not include audio content from any of those of the plurality of audio objects outside the selected subset.
  • the system may comprise an analysis section.
  • the analysis section may be configured to analyze the audio content of the audio object and determine, based on the analysis, a value indicating a level of confidence that the audio object includes dialog.
  • the system may comprise a dialog enhancement section configured to subject at least one audio object in the selected subset to dialog enhancement.
  • the dialog enhancement section may be configured to determine a degree of dialog enhancement to which to subject the at least one audio object based on the corresponding at least one determined value.
  • the system may comprise a dialog enhancement section configured to subject at least one audio object in the selected subset to dialog enhancement.
  • the selected subset may include multiple audio objects.
  • the at least one audio object to be subjected to dialog enhancement may be selected (for example by a second selection section, which may for example be comprised in the system) based on the determined values.
  • the system may comprise a second selection section configured to select at least one audio object from the selected subset based on the determined values.
  • the clustering section may be configured to performing the clustering such that the audio content from those of the plurality of audio objects outside the selected at least one audio object is included in a collection of clusters.
  • the clustering section may be configured to perform the clustering such that the at least one selected audio object (that is, the at least one audio object selected based on the determined values) is excluded from the clustering or the audio content of the at least one selected audio object is included in a cluster which does not include audio content from any of those of the plurality of audio objects outside the at least one selected audio object.
  • the system may comprise an output section configured to output a bitstream.
  • the bitstream may include the audio content, the location metadata, and the assigned value of the object type parameter.
  • Figure 1 is a perspective view of a room with an example loudspeaker setup for reproducing an audio scene.
  • Figure 10 is a top view of the room in Figure 1 .
  • Figure 1 1 is a side view of the room in Figure 1 .
  • the listener 101 sits in a sofa 102 in the middle of the room.
  • the listener 101 is watching a movie at a television screen 1 03.
  • Loudspeakers are distributed in the room to reproduce (or convey) a three-dimensional audio scene according to a 5.1 .2 loudspeaker setup.
  • a center speaker C is located at the center just below the screen 103 in front of the listener 101 .
  • a subwoofer SW for conveying low frequency effects is also arranged below the screen 103.
  • Left front LF and right front RF speakers are located in front of the listener 1 01 , but at either sides of the screen 103.
  • Left surround LS and right surround RS speakers are arranged behind the listener 101 at either side of the room.
  • Two ceiling speakers TL and TR are also provided.
  • the ceiling speakers TL and RL are arranged (or mounted) in the ceiling at either sides of the user 101 .
  • the 5.1 .2 speaker setup described with reference to figures 1 , 10 and 1 1 only serves as an example. Many other speaker setups may be employed to reproduce an audio scene. It will also be appreciated that an audio scene may be reproduced in other environments than a room with a TV. An audio scene may for example be reproduced at a cinema where loudspeakers are distributed to convey a three-dimensional audio scene while a movie is shown at the cinema screen.
  • Positions in the room may be defined via coordinates, for example Cartesian coordinates or spherical coordinates.
  • the front of the audio scene is in front of the listener 101 , where the screen 103 is and where the center speaker C is arranged.
  • the front of the audio scene may for example be defined as the plane 1 10 in which the screen 103 is arranged.
  • the back of the audio scene is behind the user 101 , for example in a plane 1 1 1 in which the left surround LS and right surround RS speakers are arranged.
  • a first coordinate 104 may define the position in the front-back direction 104
  • a second coordinate 105 may define the position in the left-right direction 105
  • a third coordinate 106 may define the vertical position 106 (or elevation) between the floor level 1 14 and the ceiling 1 13 where the ceiling speakers TL and TR are located.
  • one or more loudspeakers may be arranged below the listener level 1 12 so as to convey the impression of audio objects located below the listener 101 .
  • the third coordinate 106 may define the vertical position 106 (or elevation) between speakers located below the listener level 1 1 2 and ceiling speakers TL and TR located in the ceiling 1 13.
  • Audio objects to be reproduced by the speaker system may comprise audio content and location metadata defining a position of the audio object in the audio scene. A change in the location metadata between time frames indicates that the object is moving.
  • An impression may be conveyed that the audio object is located at its intended position in the audio scene relative to the listener 101 .
  • An example audio object 107 is shown in Figure 1 . The audio object 107 is located to the right of the listener 101 . As indicated by the arrow 108, the audio object 107 is moving backwards in the audio scene.
  • Another example audio object 109 is also shown in figure 1 . The audio object 109 is located at the front of the audio scene and does not move.
  • Audio objects including dialog may therefore be labeled accordingly, so that an audio processing system (such as an encoder, a decoder, or a renderer) receiving an audio object knows whether the audio object includes dialog.
  • an audio processing system such as an encoder, a decoder, or a renderer
  • audio processing systems receiving a collection of audio objects do not know which audio objects to subject to dialog enhancement. In such a situation, the processing system could refrain from performing dialog enhancement altogether.
  • the processing system could analyze the audio content of each of the audio objects to determine whether the respective audio objects include dialog. Examples of ways to analyze the audio content of an audio object for classifying the audio object are spectral flux speech detection and digital signal machine learning algorithms.
  • Such schemes for analyzing the audio content of audio objects are associated with relatively high computational complexity. As described below with reference to Figures 2-9, a different approach for classifying audio objects is proposed. Knowledge of whether an audio object includes dialog may also be useful in spatial coding tools aiming to reduce the total bitrate of the audio scene by reducing the objects into clusters. Such coding tools can benefit from an understanding of the object type as it allows for object isolation during the coding process to enable listener personalization during rendering, or for the switching of dialog tracks for multiple languages after spatial encoding.
  • FIG. 2 shows a system 200 according to an example embodiment.
  • an audio object has been created by a content creator.
  • the audio object comprises audio content 210 and associated location metadata 220.
  • the audio content 210 may for example be provided in the form of a pulse code modulated (PCM) file such as a WAV file.
  • the location metadata 220 may for example be provided in the form of a text file.
  • PCM pulse code modulated
  • the system 200 receives the location metadata 220 of the audio object.
  • the system 200 comprises a processing section 230 that estimates, based on the location metadata 220, whether the audio object includes dialog.
  • the processing section 230 assigns a value to an object type parameter 231 representative of the result of the estimation.
  • the object type parameter 231 may then be output by the system 200 for transmission as metadata together with the audio content 210 and the location metadata 220.
  • the object type parameter 231 may be computed in different ways.
  • the front back direction y corresponds to the front-back direction 1 04 described above with reference to Figure 1 .
  • the object type parameter is computed as a linear combination (or weighted sum) of the position y and the velocity v.
  • the coefficients k and m in equation (1 ) are weights assigned to reflect the relevance of the position y and the velocity v for estimating a confidence level of whether an audio object includes dialog. Suitable weights to apply in equation (1 ) may for example be determined via mean square fitting based on test data.
  • aspects which may be included in equation (1 ) may for example be a level of elevation of the audio object (the vertical position 106 in Figure 1 ) and/or an acceleration of the audio object (or even higher order derivatives of the location of the audio object). Audio objects located at the same level of elevation as the listener 101 (or at the same vertical position 1 12 as the listener 101 ) may be more likely to include dialog than audio objects located closer to the ceiling 1 13. Suitable weights to apply to the different aspects influencing the confidence level may for example be obtained via mean square fitting based on test data. Alternatively, a machine learning algorithm could be employed to generate a confidence level based on such different aspects.
  • the above equation (1 ) for computing the object type parameter reflects the fact that dialog audio objects in typical cinema mixes are often stationary (or slow moving) and are often located at the front 1 10 of the audio scene where the screen 103 is located.
  • the first example audio object 107 described above with reference to Figure 1 would get a low value in equation (1 ) since it is not close to the front 1 10 of the audio scene and since it is moving. This indicates that the audio object 107 is not very likely to include dialog.
  • the second example object 109 is at the front 1 10 of the audio scene and does not move. It would therefore get a higher value in equation (1 ), which indicates that is likely to include dialog.
  • Figure 3 shows a system 300 according to an example embodiment.
  • the system 300 receives location metadata 320 of an audio object, and a processing section 330 of the system 300 estimates (based on the location metadata 320) whether the audio object includes dialog.
  • the processing section 330 assigns a value to an object type parameter 331 representative of the result of the estimation.
  • the object type parameter 331 may then be output by the system 300 for transmission as metadata together with the audio content 310 and the location metadata 320 of the audio object.
  • the system 300 is a decoding system.
  • the audio content 310 and the location metadata 320 may have been encoded into a bitstream 302 via use of a multiplexer 301 , and may be retrieved from the bitstream 302 by the system 300 via use of a demultiplexer 303.
  • the decoding system 300 may for example provide its output to another system for further processing.
  • the decoding system 300 may for example provide its output to a rendering system or to an audio reproduction system.
  • the audio objects received by the system 300 could potentially have been obtained via clustering of an original set of audio objects. Such clustering may have caused stationary dialog audio objects to be included into clusters with time-varying spatial positions.
  • a received audio object including dialog may therefore move in the audio scene, even if dialog audio objects may typically be stationary. It may therefore be desirable to supplement the initial object type estimation provided by the processing section 330 with a refined object type estimation based on the audio content of the audio object, as described below with reference to Figures 4-6
  • Figure 4 shows a system 400 according to an example embodiment.
  • the system 400 receives location metadata 420 of audio objects, and a processing section 430 of the system 400 estimates (based on the location metadata 420) whether the respective audio objects includes dialog.
  • the processing section 430 assigns values to object type parameters representative of the results of the respective estimations.
  • the system 400 performs further processing via a selection section 460 and an analysis section 470.
  • the selection section 460 selects a subset of the audio objects based on the assigned values of the object type parameters.
  • the selection section 460 applies a threshold to select the subset of the audio objects.
  • each of the object type parameters indicates a confidence level for whether the associated audio object includes dialog.
  • the audio objects with a confidence level above the threshold are selected by the selection section 460. If, for example, the coefficients k and m in the above equation (1 ) both have the value 1/2 and the parameters y and v are normalized to a range between 0 and 1 , then the selection section 460 could for example employ the threshold value 1 /2, 1 /3, 1/4 or 1 /5 for the object type parameter to select audio objects.
  • the audio objects selected by the selection section 470 are then provided to the analysis section 470.
  • the analysis section 470 analyzes the audio content of the selected audio objects and determines values indicating a refined level of confidence that the respective audio objects include dialog.
  • the refined level of confidence provided by the analysis section 470 may be more reliable than the object type parameters provided by the processing section 430 since the refined confidence level is based on analysis of the actual audio content, rather than being based on the location metadata 420.
  • the processing section 430 and the selection section 460 together allow the number of audio objects to be analyzed by the analysis section 470 to be reduced, whereby the overall computational complexity is reduced. Those of the audio objects not selected by the selection section 460 may for example be assigned the refined confidence value 0.
  • the system 400 may for example be an encoding system.
  • An output section 440 may provide a bitstream 450 including the audio content 410, the location metadata 420 and the refined confidence levels for all of the audio objects.
  • Figure 5 shows a system 500 according to an example embodiment.
  • the system 500 comprises a processing section, 530, a selection section 560, an analysis section 570 and an output section 540.
  • the system 500 further comprises a second selection section 571 and a clustering section 580 which performs clustering of the audio objects before the output section includes them into a bitstream 550.
  • the second selection section 571 employs the refined confidence levels from the analysis section 570 to select which of the audio objects to be excluded from the clustering. While audio objects not including dialog may be clustered together to provide a more bitrate-efficient coding format, audio objects that are likely (or are believed) to include dialog may be too important for the overall audio experience to be clustered together with other audio objects. Such audio objects may therefore be excluded from the clustering by the second selection section 571 .
  • the clustering section 580 performs clustering and provides clusters and associated metadata to the output section 540.
  • the selected audio objects which are excluded from the clustering may for example be included in the audio bitstream together with their respective metadata.
  • Figure 6 shows a system 600 according to an example embodiment.
  • the system 600 may for example be part of a decoding system or a rendering system.
  • the system 600 receives a plurality of audio objects including audio content 610 and associated location metadata 620.
  • the audio objects received by the system 600 may for example be the result of a clustering operation performed at an encoder side to reduce the number of bits transmitted between the encoder side and the decoder side. Therefore, dialog audio objects may have been clustered together with non-dialog audio objects, so that an audio object received by the system 600 may include a combination of dialog and other audio content. Since the dialog may he been "polluted" by other audio content, it may be desirable to perform dialog enhancement to make the dialog easier for listeners to hear and/or distinguish from the rest of the audio content.
  • a processing section 630 classifies the audio objects based on the location metadata. More specifically, the processing section 630 estimates, based on the location metadata 620, whether the audio object includes dialog, and assigns a value to an object type parameter based on the result of the estimation.
  • the value of the object type parameter is compared to a threshold in a comparator 661 such that only the audio objects most likely to contain dialog get enhanced.
  • the output of the comparator 661 is a Boolean decision whether to allow further processing of the audio object, controlled by a switch 662.
  • the comparator 661 and the switch 662 together act as a selection section similar to the selection sections 460 and 560 described above with reference to figures 4 and 5.
  • an audio object qualifies for further processing that is, if the object type parameter is above the threshold employed by the comparator 661 , the audio content 610 of that audio object is analyzed in an analysis section 670.
  • Figure 7 is a flow chart of a method 700 according to an embodiment.
  • the method 700 may for example be performed by any of the systems 200-600 described above with reference to figures 2-6.
  • the method 700 comprises receiving 701 location metadata associated with an audio object, wherein the location metadata defines a position of the audio object in an audio scene, estimating 702, based on the location metadata, whether the audio object includes dialog, and assigning 703 a value to an object type parameter representative of a result of the estimation.
  • Figure 9 is a flow chart of a method 900 according to an embodiment.
  • the method 900 may for example be performed by the system 500, described above with reference to Figure 5.
  • the method 900 may for example include the same steps 801 - 806 as the method 800 described above with reference to figure 8, but the last step may relate to clustering 907 instead of dialog enhancement 807. More specifically, the clustering 907 may be performed based on the refined object type parameter values determined at the step 806. Audio objects with high enough refined object type parameter values may be excluded from the clustering. In other words, those audio objects believed to include dialog are excluded from the clustering, while the other audio objects are clustered together into one or more clusters.
  • the devices and methods disclosed hereinabove may be implemented as software, firmware, hardware or a combination thereof.
  • the division of tasks between functional units referred to in the above description does not necessarily correspond to the division into physical units; to the contrary, one physical component may have multiple functionalities, and one task may be carried out by several physical components in cooperation.
  • Certain components or all components may be implemented as software executed by a digital signal processor or microprocessor, or be implemented as hardware or as an application-specific integrated circuit.
  • Such software may be distributed on computer readable media, which may comprise computer storage media (or non-transitory media) and communication media (or transitory media).
  • Computer storage media includes both volatile and nonvolatile, removable and non-removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data.
  • Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disks (DVD) or other optical disk storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other medium which can be used to store the desired information and which can be accessed by a computer.
  • communication media typically embodies computer readable instructions, data structures, program modules or other data in a modulated data signal such as a carrier wave or other transport mechanism and includes any information delivery media.
  • EEEs enumerated example embodiments
  • EEE3 The method of any of the preceding EEEs, wherein the estimation is performed based on a position of the audio object in a front-back direction (104) of the audio scene, the position in the front-back direction being defined by the location metadata.
  • EEE4 The method of EEE 3, wherein estimating whether the audio object includes dialog comprises:
  • EEE9 The method of EEE 8, wherein estimating whether the audio object includes dialog comprises:
  • the first level of elevation corresponds to a floor level (1 14) of the audio scene or a vertical position (1 12) of an intended listener (101 ).
  • each of the received audio objects including audio content (610) and location metadata (620), wherein the location metadata of an audio object defines a position of that audio object in an audio scene; estimating (802), based on the location metadata of the respective audio objects, whether the respective audio objects include dialog;
  • EEE12. The method of any of EEEs 10-1 1 , further comprising:
  • EEE13 The method (900) of any of EEEs 10-12, further comprising performing clustering (907) such that the audio content from those of the plurality of audio objects outside the selected subset is included in a collection of clusters and such that:
  • At least one audio object of the selected subset is excluded from the clustering
  • the audio content of at least one audio object of the selected subset is included in a cluster which does not include audio content from any of those of the plurality of audio objects outside the selected subset.
  • EEE14 The method of any of EEEs 10-13, further comprising, for each of the one or more audio objects in the selected subset:
  • EEE15 The method (800) of EEE 14, comprising:
  • EEE16 The method (800) of any of EEEs 14-15, wherein the selected subset includes multiple audio objects, the method comprising:
  • clustering is performed such that:
  • the at least one selected audio object is excluded from the clustering; or the audio content of the at least one selected audio object is included in a cluster which does not include audio content from any of those of the plurality of audio objects outside the at least one selected audio object.
  • EEE18 A computer program product comprising a computer-readable medium with instructions for performing the method of any of EEEs 1 -17.
  • a processing section configured to estimate, based on the location metadata, whether the audio object includes dialog, and to assign a value to an object type parameter (231 ) representative of a result of the estimation.
  • a selection section (460, 560, 661 , 662) configured to select a subset of the plurality of audio objects based on the assigned values of the object type
  • EEE21 The system of EEE 20, further comprising:
  • dialog enhancement section (690) configured to subject at least one audio object in the selected subset to dialog enhancement.
  • At least one audio object of the selected subset is excluded from the clustering
  • the audio content of at least one audio object of the selected subset is included in a cluster which does not include audio content from any of those of the plurality of audio objects outside the selected subset.
  • EEE23 The system of any of EEEs 20-22, further comprising an analysis section (470, 570, 670) configured to, for each of the one or more audio objects in the selected subset:
  • EEE24 The system of EEE 23, wherein the system comprises:
  • dialog enhancement section (690) configured to subject at least one audio object in the selected subset to dialog enhancement, wherein the dialog
  • EEE25 The system of any of EEEs 23-24, wherein the system comprises: a dialog enhancement section (690) configured to subject at least one audio object in the selected subset to dialog enhancement, wherein the selected subset includes multiple audio objects, and wherein the at least one audio object to be subjected to dialog enhancement is selected based on the determined values.
  • a dialog enhancement section (690) configured to subject at least one audio object in the selected subset to dialog enhancement, wherein the selected subset includes multiple audio objects, and wherein the at least one audio object to be subjected to dialog enhancement is selected based on the determined values.
  • a second selection section configured to select at least one audio object from the selected subset based on the determined values
  • the at least one selected audio object is excluded from the clustering; or the audio content of the at least one selected audio object is included in a cluster which does not include audio content from any of those of the plurality of audio objects outside the at least one selected audio object.

Landscapes

  • Engineering & Computer Science (AREA)
  • Computational Linguistics (AREA)
  • Signal Processing (AREA)
  • Health & Medical Sciences (AREA)
  • Audiology, Speech & Language Pathology (AREA)
  • Human Computer Interaction (AREA)
  • Physics & Mathematics (AREA)
  • Acoustics & Sound (AREA)
  • Multimedia (AREA)
  • Stereophonic System (AREA)

Abstract

La présente invention concerne des procédés (700, 800, 900), des systèmes (200, 300, 400, 500, 600) et des produits de programme informatique. Des métadonnées de localisation (620) associées à un objet audio sont reçues (801). Les métadonnées de localisation définissent une position de l'objet audio dans une scène audio. Il est estimé (630, 802), sur la base des métadonnées de localisation, si l'objet audio comprend un dialogue. Une valeur représentative d'un résultat de l'estimation est attribuée (803) à un paramètre de type d'objet (231). Dans certains modes de réalisation exemplaires, des objets audio sont sélectionnés (661, 662, 804) sur la base de valeurs de leurs paramètres de type d'objet respectifs. Dans certains modes de réalisation exemplaires, au moins un des objets audio sélectionnés est soumis à une accentuation de dialogue (690, 807).
PCT/US2018/043980 2017-08-01 2018-07-26 Classification d'objet audio sur la base de métadonnées de localisation WO2019027812A1 (fr)

Priority Applications (3)

Application Number Priority Date Filing Date Title
CN201880049177.7A CN110998724B (zh) 2017-08-01 2018-07-26 基于位置元数据的音频对象分类
EP18747091.9A EP3662470B1 (fr) 2017-08-01 2018-07-26 Classification d'objet audio basée sur des métadonnées de localisation
US16/636,241 US11386913B2 (en) 2017-08-01 2018-07-26 Audio object classification based on location metadata

Applications Claiming Priority (4)

Application Number Priority Date Filing Date Title
US201762539599P 2017-08-01 2017-08-01
EP17184244.6 2017-08-01
US62/539,599 2017-08-01
EP17184244 2017-08-01

Publications (1)

Publication Number Publication Date
WO2019027812A1 true WO2019027812A1 (fr) 2019-02-07

Family

ID=59506166

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/US2018/043980 WO2019027812A1 (fr) 2017-08-01 2018-07-26 Classification d'objet audio sur la base de métadonnées de localisation

Country Status (1)

Country Link
WO (1) WO2019027812A1 (fr)

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20100121634A1 (en) 2007-02-26 2010-05-13 Dolby Laboratories Licensing Corporation Speech Enhancement in Entertainment Audio
US20150332680A1 (en) 2012-12-21 2015-11-19 Dolby Laboratories Licensing Corporation Object Clustering for Rendering Object-Based Audio Content Based on Perceptual Criteria
US20160078879A1 (en) 2013-03-26 2016-03-17 Dolby Laboratories Licensing Corporation Apparatuses and Methods for Audio Classifying and Processing
WO2016172111A1 (fr) * 2015-04-20 2016-10-27 Dolby Laboratories Licensing Corporation Traitement de données audio pour compenser une perte auditive partielle ou un environnement auditif indésirable
US20170098452A1 (en) * 2015-10-02 2017-04-06 Dts, Inc. Method and system for audio processing of dialog, music, effect and height objects

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20100121634A1 (en) 2007-02-26 2010-05-13 Dolby Laboratories Licensing Corporation Speech Enhancement in Entertainment Audio
US20150332680A1 (en) 2012-12-21 2015-11-19 Dolby Laboratories Licensing Corporation Object Clustering for Rendering Object-Based Audio Content Based on Perceptual Criteria
US20160078879A1 (en) 2013-03-26 2016-03-17 Dolby Laboratories Licensing Corporation Apparatuses and Methods for Audio Classifying and Processing
WO2016172111A1 (fr) * 2015-04-20 2016-10-27 Dolby Laboratories Licensing Corporation Traitement de données audio pour compenser une perte auditive partielle ou un environnement auditif indésirable
US20170098452A1 (en) * 2015-10-02 2017-04-06 Dts, Inc. Method and system for audio processing of dialog, music, effect and height objects

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
KUBA LOPATKA ET AL: "Improving listeners' experience for movie playback through enhancing dialogue clarity in soundtracks", DIGITAL SIGNAL PROCESSING., vol. 48, 1 January 2016 (2016-01-01), US, pages 40 - 49, XP055446566, ISSN: 1051-2004, DOI: 10.1016/j.dsp.2015.08.015 *

Similar Documents

Publication Publication Date Title
US11064310B2 (en) Method, apparatus or systems for processing audio objects
EP2936485B1 (fr) Groupage d'objets pour le rendu du contenu des objets audio sur la base des critères perceptuels
US9282417B2 (en) Spatial sound reproduction
US10638246B2 (en) Audio object extraction with sub-band object probability estimation
CN104584121B (zh) 音频水印的缩混补偿方法、系统及装置
AU2006233504B2 (en) Apparatus and method for generating multi-channel synthesizer control signal and apparatus and method for multi-channel synthesizing
JP2014515906A (ja) オーディオをアップミックスして３ｄオーディオを生成する方法とシステム［関連出願との相互参照］この出願は、２０１１年４月１８日に出願された米国特許仮出願第６１／４７６，３９５号の優先権を主張するものである。この文献はここにその全体を参照援用する。
WO2020008112A1 (fr) Signalisation et synthèse de rapport énergétique
US11386913B2 (en) Audio object classification based on location metadata
WO2016049106A1 (fr) Introduction d'objets sonores dans un signal audio à mixage réducteur
WO2019027812A1 (fr) Classification d'objet audio sur la base de métadonnées de localisation
JP2023514121A (ja) ビデオ情報に基づく空間オーディオ拡張

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 18747091

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

ENP Entry into the national phase

Ref document number: 2018747091

Country of ref document: EP

Effective date: 20200302