Nokia Corporation

Nokia Technologies Oy

Core Wireless Licensing S.a.r.l.

Victor Popa

Jani Kristian Nurminen

Jilei Tian

Elina Elisabet Helander

Antti Johannes Eronen

Jussi Artturi Leppänen

Igor Danilo Diego Curcio

Tuomas Virtanen

Katariina Jutta Mahkonen

Jilie Tian

An apparatus comprises a dereverberation module for generating a dereverberated audio signal based on an original audio signal containing reverberation, and an audio-analysis module for generating audio analysis data based on audio analysis of the original audio signal and audio analysis of the dereverberated audio signal.

Audio signal analysis

An apparatus for providing text independent voice conversion may include a first voice conversion model and a second voice conversion model. The first voice conversion model may be trained with respect to conversion of training source speech to synthetic speech corresponding to the training source speech. The second voice conversion model may be trained with respect to conversion to training target speech from synthetic speech corresponding to the training target speech. An output of the first voice conversion model may be communicated to the second voice conversion model to process source speech input into the first voice conversion model into target speech corresponding to the source speech as the output of the second voice conversion model.

Method, apparatus and computer program product for providing text independent voice conversion

A hybrid approach is described for combining frequency warping and Gaussian Mixture Modeling (GMM) to achieve better speaker identity and speech quality. To train the voice conversion GMM model, line spectral frequency and other features are extracted from a set of source sounds to generate a source feature vector and from a set of target sounds to generate a target feature vector. The GMM model is estimated based on the aligned source feature vector and the target feature vector. A mixture specific warping function is generated each set of mixture mean pairs of the GMM model, and a warping function is generated based on a weighting of each of the mixture specific warping functions. The warping function can be used to convert sounds received from a source speaker to approximate speech of a target speaker.

Hybrid approach in voice conversion

An apparatus for providing voice conversion using temporal dynamic features includes a feature extractor and a transformation element. The feature extractor may be configured to extract dynamic feature vectors from source speech. The transformation element may be in communication with the feature extractor and configured to apply a first conversion function to a signal including the extracted dynamic feature vectors to produce converted dynamic feature vectors. The first conversion function may have been trained using at least dynamic feature data associated with training source speech and training target speech. The transformation element may be further configured to produce converted speech based on an output of applying the first conversion function.

Method, apparatus and computer program product for providing voice conversion using temporal dynamic features

It may be desirable to provide a way to collect high quality speech training data without undue burden to the user. Speech training data may be collected during normal usage of a device. In this way, the collection of speech training data may be effectively transparent to the user, without the need for a distinct collection mode from the user's point of view. For example, where the device is or includes a phone (such as a cellular phone), when the user makes or receives a phone call to/from another party, speech training data may be automatically collected from one or both of the parties during the phone call.

Voice conversion training and data collection

An apparatus for providing data clustering and mode selection includes a training element and a transformation element. The training element is configured to receive a first training data set, a second training data set and auxiliary data extracted from the same material as the first training data set. The training element is also configured to train a classifier to group the first training data set into M clusters based on the auxiliary data and the first training data set and train M processing schemes corresponding to the M clusters for transforming the first training data set into the second training data set. The transformation element is in communication with the training element and is configured to cluster the second training data set into M clusters based on features associated with the second training data set.

Method, apparatus, mobile terminal and computer program product for providing data clustering and mode selection

Systems and methods are provided for performing soft alignment in Gaussian mixture model (GMM) based and other vector transformations. Soft alignment may assign alignment probabilities to source and target feature vector pairs. The vector pairs and associated probabilities may then be used calculate a conversion function, for example, by computing GMM training parameters from the joint vectors and alignment probabilities to create a voice conversion function for converting speech sounds from a source speaker to a target speaker.

Soft alignment based on a probability of time alignment

An apparatus for providing efficient evaluation of feature transformation includes a training module and a transformation module. The training module is configured to train a Gaussian mixture model (GMM) using training source data and training target data. The transformation module is in communication with the training module. The transformation module is configured to produce a conversion function in response to the training of the GMM. The training module is further configured to determine a quality of the conversion function prior to use of the conversion function by calculating a trace measurement of the GMM.

Growing community of inventors

Tampere, Finland

Victor Popa