Adobe, Inc.

Apple Inc.

Northwestern University

Nicholas J Bryan

Paris Smaragdis

Vasu Iyengar

Zeyu Jin

Justin Salamon

Dingzeyu Li

Timothy Langlois

Jongpil Lee

Maxwell Morrison

Zhenyu Tang

Juan Pablo Caceres Chomali

Aram M Lindahl

Oliver Wang

Gautham J Mysore

Qing Yang

Bryan A Pardo

Marco Antonio Martinez Ramirez

Lucas Rencker

Stylianos Ioannis Mimilakis

Yu Wang

Embodiments are disclosed for performing a using a neural network to optimize filter weights of an adaptive filter. In particular, in one or more embodiments, the disclosed systems and methods comprise receiving, by a filter, an input audio signal, wherein the input audio signal is a far-end audio signal, the filter including a transfer function with adaptable filter weights, generating a response audio signal modeling the input audio signal passing through the acoustic environment, receiving a target response signal, including the input audio signal and near-end audio signals, calculating an adaptive filter loss, generating, by a trained recurrent neural network, a filter weight update using the calculated adaptive filter loss, updating the adaptable filter weights of the transfer function to create an updated transfer function, generating an updated response audio signal based on the updated transfer function, and providing the updated response audio signal as an output audio signal.

Meta-learning for adaptive filters

Methods for modifying audio data include operations for accessing audio data having a first prosody, receiving a target prosody differing from the first prosody, and computing acoustic features representing samples. Computing respective acoustic features for a sample includes computing a pitch feature as a quantized pitch value of the sample by assigning a pitch value, of the target prosody or the audio data, to at least one of a set of pitch bins having equal widths in cents. Computing the respective acoustic features further includes computing a periodicity feature from the audio data. The respective acoustic features for the sample include the pitch feature, the periodicity feature, and other acoustic features. A neural vocoder is applied to the acoustic features to pitch-shift and time-stretch the audio data from the first prosody toward the target prosody.

Neural pitch-shifting and time-stretching

Embodiments are disclosed for determining an answer to a query associated with a graphical representation of data. In particular, in one or more embodiments, the disclosed systems and methods comprise receiving an input including an unprocessed audio sequence and a request to perform an audio signal processing effect on the unprocessed audio sequence. The one or more embodiments further include analyzing, by a deep encoder, the unprocessed audio sequence to determine parameters for processing the unprocessed audio sequence. The one or more embodiments further include sending the unprocessed audio sequence and the parameters to one or more audio signal processing effects plugins to perform the requested audio signal processing effect using the parameters and outputting a processed audio sequence after processing of the unprocessed audio sequence using the parameters of the one or more audio signal processing effects plugins.

Deep encoder for performing audio processing

Methods are performed by one or more processing devices for correcting prosody in audio data. A method includes operations for accessing subject audio data in an audio edit region of the audio data. The subject audio data in the audio edit region potentially lacks prosodic continuity with unedited audio data in an unedited audio portion of the audio data. The operations further include predicting, based on a context of the unedited audio data, phoneme durations including a respective phoneme duration of each phoneme in the unedited audio data. The operations further include predicting, based on the context of the unedited audio data, a pitch contour comprising at least one respective pitch value of each phoneme in the unedited audio data. Additionally, the operations include correcting prosody of the subject audio data in the audio edit region by applying the phoneme durations and the pitch contour to the subject audio data.

Context-aware prosody correction of edited speech

Methods, systems, and non-transitory computer readable storage media are disclosed for rendering scene-aware audio based on acoustic properties of a user environment. For example, the disclosed system can use neural networks to analyze an audio recording to predict environment equalizations and reverberation decay times of the user environment without using a captured impulse response of the user environment. Additionally, the disclosed system can use the predicted reverberation decay times with an audio simulation of the user environment to optimize material parameters for the user environment. The disclosed system can then generate an audio sample that includes scene-aware acoustic properties based on the predicted environment equalizations, material parameters, and an environment geometry of the user environment. Furthermore, the disclosed system can augment training data for training the neural networks using frequency-dependent equalization information associated with measured and synthetic impulse responses.

Generating scene-aware audio using a neural network-based acoustic analysis

In implementations of searching for music, a music search system can receive a music search request that includes a music file including music content. The music search system can also receive a selected musical attribute from a plurality of musical attributes. The music search system includes a music search application that can generate musical features of the music content, where a respective one or more of the musical features correspond to a respective one of the musical attributes. The music search application can then compare the musical features that correspond to the selected musical attribute to audio features of audio files, and determine similar audio files to the music file based on the comparison of the musical features to the audio features of the audio files.

Searching for music

Certain embodiments involve techniques for automatically identifying sounds in an audio recording that match a selected sound. An audio search and editing system receives the audio recording and preprocesses the audio recording into audio portions. The audio portions are provided as a query to the neural network that includes a trained embedding model used to analyze the audio portions in view of the selected sound to estimate feature vectors. The audio search and editing system compares the feature vectors for the audio portions against the feature vector for the selected sound and the feature vector for the negative samples to generate an audio score that is a numerical representation of the level of similarity between the audio portion and the selected sound and uses the audio scores to classify the audio portions into a first class of matching sounds and a second class of non-matching sounds.

Automated sound matching within an audio recording

Rendering scene-aware audio using neural network-based acoustic analysis

A device implementing a system for mitigating noise includes at least one processor configured to receive a first audio signal corresponding to a first microphone, and determine whether wind noise is present based at least in part on the first audio signal. The processor is configured to select, based on the determining, a second audio signal from between second and third microphones. The second microphone is disposed at a location that experiences less echo coupling when the device is in a particular orientation with respect to a user. The third microphone is disposed at another location that experiences less wind noise. The processor is configured to determine voice and noise reference values based on the first and the selected second audio signals, and perform noise suppression with respect to at least one of the first or the selected second audio signal, based on the voice or the noise reference value.

Mitigating noise in audio signals

One example method involves operations for receiving input to transform audio to a target style. Operations further include providing the audio to a predictive model trained to transform the audio into produced audio. Training the predictive model includes accessing representations of audios and unpaired audios. Further, training includes generating feature embeddings by extracting features from representations of an audio and an unpaired audio. The unpaired audio includes a reference production style, and the feature embeddings correspond to their representations. Training further includes generating a feature vector by comparing the feature embeddings using a comparison model. Further, training includes computing prediction parameters using a learned function. The prediction parameters can transform the feature vector into the reference style. Training further includes updating the predictive model with the prediction parameters. In addition, operations include generating the produced audio by modifying audio effects of the audio into the target style.

Audio production assistant for style transfers of audio recordings using one-shot parametric predictions

The disclosure describes one or more embodiments of an impulse response system that generates accurate and realistic synthetic impulse responses. For example, given an acoustic impulse response, the impulse response system can generate one or more synthetic impulse responses that modify the direct-to-reverberant ratio (DRR) of the acoustic impulse response. As another example, the impulse response system can generate one or more synthetic impulse responses that modify the reverberation time (e.g., T60) of the acoustic impulse response. Further, utilizing the synthetic impulse responses, the impulse response system can perform a variety of functions to improve a digital audio recording or acoustic measurement or prediction model.

Generating synthetic acoustic impulse responses from an acoustic impulse response

A method for controlling a speech enhancement process in a far-end device, while engaged in a voice or video telephony communication session over a communication link with a near-end device. A near-end user speech signal is produced, using a microphone to pick up speech of a near-end user, and is analyzed by an automatic speech recognizer (ASR) without being triggered by an ASR trigger phrase or button. The recognized words are compared to a library of phrases to select a matching phrase, where each phrase is associated with a message that represents an audio signal processing operation. The message associated with the matching phrase is sent to the far-end device, which is used to configure the far-end device to adjust the speech enhancement process that produces the far-end speech signal. Other embodiments are also described.

Transparent near-end user control over far-end speech enhancement processing

Signals are received from audio pickup channels that contain signals from multiple sound sources. The audio pickup channels may include one or more microphones and one or more accelerometers. Signals representative of multiple sound sources are generated using a blind source separation algorithm. It is then determined which of those signals is deemed to be a voice signal and which is deemed to be a noise signal. The output noise signal may be scaled to match a level of the output voice signal, and a clean speech signal is generated based on the output voice signal and the scaled noise signal. Other aspects are described.

Speech enhancement for an electronic device

System of noise reduction for mobile devices includes blind source separator (BSS) and noise suppressor. BSS receives signals from at least two audio pickup channels. BSS includes sound source separator, voice source detector, equalizer, and auto-disabler. Sound source separator generates signals representing first sound source and second sound source based on signals from the first and the second channels. Voice source detector determines whether the signals representing the first and second sound sources are voice signal or noise signal, respectively. Equalizer scales noise signal to match a level of the voice signal, and generates scaled noise signal. Auto-disabler determines whether to disable BSS. Auto-disabler outputs signals from the at least two audio pickup channels when the BSS is disabled and outputs the voice signal and the scaled noise signal when the BSS is not disabled. Noise suppressor generates clean signal based on outputs from auto-disabler. Other embodiments are also described.

Growing community of inventors

San Francisco, CA, United States of America

Nicholas J Bryan