Abstract: Techniques are disclosed for automatic detection of dense ornamentation in music. Input data representing a piece of digitally encoded music in a time domain is converted into a spectrogram representing time-frequency coefficients in a frequency domain. The spectrogram includes column vectors of the time-frequency coefficients that correspond to time periods spanning different portions of the piece of music. A one-dimensional onset detection array is calculated based on a subset of the column vectors. Using the spectrogram and the onset detection array, a two-dimensional self-similarity matrix (SSM) is calculated based on pair-wise comparisons of elements in the onset detection array. As a result, an irregular pattern score representing the presence of dense ornamentation in the piece of music can be calculated based on a magnitude difference between a beat pattern in the music and each column of the slim SSM.

Abstract: Sound feature priority alignment techniques are described. In one or more implementations, features of sound data are identified from a plurality of recordings. Values are calculated for frames of the sound data from the plurality of recordings. The values are based on similarity of the frames of the sound data from the plurality of recordings to each other, the similarity based on the identified features and a priority that is assigned based on the identified features of respective frames. The sound data from the plurality of recordings is then aligned based at least in part on the calculated values.

Abstract: Pattern matching of sound data using hashing is described. In one or more implementations, a query formed from one or more spectrograms of sound data is hashed and used to locate one or more labels in a database of sound signals. Each of the labels is located using a hash of an entry in the database. At least one of the located one or more labels is chosen as corresponding to the query.

Abstract: Sound decomposition models are described. In one or more implementations, a plurality of individual models is generated for respective ones of a plurality of sound sources. The plurality of models is collected to form a universal audio model that is configured to support sound decomposition of sound data through use of one or more of the models. The plurality of models is not generated using a sound source that originated at least a portion of the sound data.

Abstract: Sound enhancement techniques through dereverberation are described. In one or more implementations, a method is described of enhancing sound data through removal of reverberation from the sound data by one or more computing devices. The method includes obtaining a model that describes primary sound data that is to be utilized as a prior that assumes no prior knowledge about specifics of the sound data from which the reverberation is to be removed. A reverberation kernel is computed having parameters that, when applied to the model that describes the primary sound data, corresponds to the sound data from which the reverberation is to be removed. The reverberation is removed from the sound data using the reverberation kernel.

Abstract: Acoustic matching and splicing of sound tracks is described. In one or more implementations, a method to acoustically match and splice first and second sound tracks by one or more computing devices is described. The method includes source separating the first and second sound tracks into first track primary and background sound data and second track primary and background sound data. Features extracted from the first and second primary sound data are matched, one to another, to generate first and second primary matching masks. Features extracted from the first and second background sound data are matched, one to another, to generate first and second background matching masks, which are applied to respective separated sound data. The applied first track primary and background sound data and the applied second track primary and background sound data are spliced to generate a spliced sound track.

Abstract: Sound alignment techniques that employ timing information are described. In one or more implementations, features and timing information of sound data generated from a first sound signal are identified and used to identify features of sound data generated from a second sound signal. The identified features may then be utilized to align portions of the sound data from the first and second sound signals to each other.

Abstract: Multichannel sound source identification and location techniques are described. In one or more implementations, source separation is performed using a collaborative technique for a plurality of sound data that was captured by respective ones of a plurality of sound capture devices of an audio scene. The source separation is performed by recognizing spectral and temporal aspects from the plurality of sound data and sharing the recognized spectral and temporal aspects, one with another, to identify one or more sound sources in the audio scene. A relative position of the identified one or more sounds sources to the plurality of sound capture devices is determined based on the source separation.

Abstract: Joint sound model generation techniques are described. In one or more implementations, a plurality of models of sound data received from a plurality of different sound scenes are jointly generated. The joint generating includes learning information as part of generating a first said model of sound data from a first one of the sound scenes and sharing the learned information for use in generating a second one of the models of sound data from a second one of the sound scenes.

Abstract: Performance metric based stopping criteria for iterative algorithm techniques are described. In one or more implementations, a training dataset is processed by one or more computing devices using an iterative algorithm having a cost function. The processing includes, for a plurality of iterations of the iterative algorithm, computing a cost for the iterative algorithm using the cost function and a value for each of a plurality of performance metrics that are usable to infer accuracy of the iterative algorithm for a respective one of the iterations. Responsive to the processing, a particular one of the plurality of iterations is identified as a stopping criterion based at least in part on the computed values for the plurality of performance metrics and the stopping criterion is output to configure the iterative algorithm to use the stopping criterion for subsequent processing of data by the iterative algorithm.

Abstract: Sound alignment user interface techniques are described. In one or more implementations, a user interface is output having a first representation of sound data generated from a first sound signal and a second representation of sound data generated from a second sound signal. One or more inputs are received, via interaction with the user interface, that indicate that a first point in time in the first representation corresponds to a second point in time in the second representation. Aligned sound data is generated from the sound data from the first and second sound signals based at least in part on correspondence of the first point in time in the sound data generated from the first sound signal to the second point in time in the sound data generated from the second sound signal.

Abstract: A sound mixture may be received that includes a plurality of sources. A model may be received that includes a dictionary of spectral basis vectors for the plurality of sources. A weight may be estimated for each of the plurality of sources in the sound mixture based on the model. In some examples, such weight estimation may be performed using a source separation technique without actually separating the sources.

Abstract: Sound processing techniques using recurrent neural networks are described. In one or more implementations, temporal dependencies are captured in sound data that are modeled through use of a recurrent neural network (RNN). The captured temporal dependencies are employed as part of feature extraction performed using nonnegative matrix factorization (NMF). One or more sound processing techniques are performed on the sound data based at least in part on the feature extraction.

Abstract: Multichannel sound source identification and location techniques are described. In one or more implementations, source separation is performed using a collaborative technique for a plurality of sound data that was captured by respective ones of a plurality of sound capture devices of an audio scene. The source separation is performed by recognizing spectral and temporal aspects from the plurality of sound data and sharing the recognized spectral and temporal aspects, one with another, to identify one or more sound sources in the audio scene. A relative position of the identified one or more sounds sources to the plurality of sound capture devices is determined based on the source separation.

Abstract: Sound processing using a product-of-filters model is described. In one or more implementations, a model is formed by one or more computing devices for a time frame of sound data as a product of filters. The model is utilized by the one or more computing devices to perform one or more sound processing techniques on the time frame of the sound data.

Abstract: Pattern identification using convolution is described. In one or more implementations, a representation of a pattern is obtained that is described using data points that include frequency coordinates, time coordinates, and energy values. An identification is made as to whether sound data described using irregularly positioned data points includes the pattern, the identifying including use of a convolution of the frequency or time coordinates to determine correspondence with the representation of the pattern.

Abstract: Pattern matching of sound data using hashing is described. In one or more implementations, a query formed from one or more spectrograms of sound data is hashed and used to locate one or more labels in a database of sound signals. Each of the labels is located using a hash of an entry in the database. At least one of the located one or more labels is chosen as corresponding to the query.

Abstract: A sound mixture may be received that includes a plurality of sources. A model may be received for one of the source that includes a dictionary of spectral basis vectors corresponding to that one source. At least one feature of the one source in the sound mixture may be estimated based on the model. In some examples, the estimation may be constrained according to temporal data.

Abstract: Joint sound model generation techniques are described. In one or more implementations, a plurality of models of sound data received from a plurality of different sound scenes are jointly generated. The joint generating includes learning information as part of generating a first said model of sound data from a first one of the sound scenes and sharing the learned information for use in generating a second one of the models of sound data from a second one of the sound scenes.

Abstract: A system and method are described for selecting a target sound object from a sound mixture. In embodiments, a sound mixture comprises a plurality of sound objects superimposed in time. A user can select one of these sound objects by providing reference audio data corresponding to a reference sound object. The system analyzes the audio data and the reference audio data to identify a portion of the audio data corresponding to a target sound object in the mixture that is most similar to the reference sound object. The analysis may include decomposing the reference audio data into a plurality of reference components and the sound mixture into a plurality of components guided by the reference components. The target sound object can be re-synthesized from the target components.