Hakan Erdogan

Hakan Erdogan, Scott Wisdom, Xuankai Chang et al. · deepmind

We present TokenSplit, a speech separation model that acts on discrete token sequences. The model is trained on multiple tasks simultaneously: separate and transcribe each speech source, and generate speech from text. The model operates on transcripts and audio token sequences and achieves multiple tasks through masking of inputs. The model is a sequence-to-sequence encoder-decoder model that uses the Transformer architecture. We also present a "refinement" version of the model that predicts enhanced audio tokens from the audio tokens of speech separated by a conventional separation model. Using both objective metrics and subjective MUSHRA listening tests, we show that our model achieves excellent performance in terms of separation, both with or without transcript conditioning. We also measure the automatic speech recognition (ASR) performance and provide audio samples of speech synthesis to demonstrate the additional utility of our model.

Yang Yang, Shao-Fu Shih, Hakan Erdogan et al.

High quality speech capture has been widely studied for both voice communication and human computer interface reasons. To improve the capture performance, we can often find multi-microphone speech enhancement techniques deployed on various devices. Multi-microphone speech enhancement problem is often decomposed into two decoupled steps: a beamformer that provides spatial filtering and a single-channel speech enhancement model that cleans up the beamformer output. In this work, we propose a speech enhancement solution that takes both the raw microphone and beamformer outputs as the input for an ML model. We devise a simple yet effective training scheme that allows the model to learn from the cues of the beamformer by contrasting the two inputs and greatly boost its capability in spatial rejection, while conducting the general tasks of denoising and dereverberation. The proposed solution takes advantage of classical spatial filtering algorithms instead of competing with them. By design, the beamformer module then could be selected separately and does not require a large amount of data to be optimized for a given form factor, and the network model can be considered as a standalone module which is highly transferable independently from the microphone array. We name the ML module in our solution as GSENet, short for Guided Speech Enhancement Network. We demonstrate its effectiveness on real world data collected on multi-microphone devices in terms of the suppression of noise and interfering speech.

Scott Wisdom, Hakan Erdogan, Daniel Ellis et al.

We introduce the Free Universal Sound Separation (FUSS) dataset, a new corpus for experiments in separating mixtures of an unknown number of sounds from an open domain of sound types. The dataset consists of 23 hours of single-source audio data drawn from 357 classes, which are used to create mixtures of one to four sources. To simulate reverberation, an acoustic room simulator is used to generate impulse responses of box shaped rooms with frequency-dependent reflective walls. Additional open-source data augmentation tools are also provided to produce new mixtures with different combinations of sources and room simulations. Finally, we introduce an open-source baseline separation model, based on an improved time-domain convolutional network (TDCN++), that can separate a variable number of sources in a mixture. This model achieves 9.8 dB of scale-invariant signal-to-noise ratio improvement (SI-SNRi) on mixtures with two to four sources, while reconstructing single-source inputs with 35.5 dB absolute SI-SNR. We hope this dataset will lower the barrier to new research and allow for fast iteration and application of novel techniques from other machine learning domains to the sound separation challenge.

Lyria Team, Antoine Caillon, Brian McWilliams et al.

We introduce a new class of generative models for music called live music models that produce a continuous stream of music in real-time with synchronized user control. We release Magenta RealTime, an open-weights live music model that can be steered using text or audio prompts to control acoustic style. On automatic metrics of music quality, Magenta RealTime outperforms other open-weights music generation models, despite using fewer parameters and offering first-of-its-kind live generation capabilities. We also release Lyria RealTime, an API-based model with extended controls, offering access to our most powerful model with wide prompt coverage. These models demonstrate a new paradigm for AI-assisted music creation that emphasizes human-in-the-loop interaction for live music performance.

Daniel P. W. Ellis, Eduardo Fonseca, Ron J. Weiss et al. · deepmind

Editing complex real-world sound scenes is difficult because individual sound sources overlap in time. Generative models can fill-in missing or corrupted details based on their strong prior understanding of the data domain. We present a system for editing individual sound events within complex scenes able to delete, insert, and enhance individual sound events based on textual edit descriptions (e.g., ``enhance Door'') and a graphical representation of the event timing derived from an ``event roll'' transcription. We present an encoder-decoder transformer working on SoundStream representations, trained on synthetic (input, desired output) audio example pairs formed by adding isolated sound events to dense, real-world backgrounds. Evaluation reveals the importance of each part of the edit descriptions -- action, class, timing. Our work demonstrates ``recomposition'' is an important and practical application.

Tuochao Chen, D Shin, Hakan Erdogan et al.

This paper introduces SoundSculpt, a neural network designed to extract target sound fields from ambisonic recordings. SoundSculpt employs an ambisonic-in-ambisonic-out architecture and is conditioned on both spatial information (e.g., target direction obtained by pointing at an immersive video) and semantic embeddings (e.g., derived from image segmentation and captioning). Trained and evaluated on synthetic and real ambisonic mixtures, SoundSculpt demonstrates superior performance compared to various signal processing baselines. Our results further reveal that while spatial conditioning alone can be effective, the combination of spatial and semantic information is beneficial in scenarios where there are secondary sound sources spatially close to the target. Additionally, we compare two different semantic embeddings derived from a text description of the target sound using text encoders.

Yang Yang, George Sung, Shao-Fu Shih et al.

We propose a neural network model that can separate target speech sources from interfering sources at different angular regions using two microphones. The model is trained with simulated room impulse responses (RIRs) using omni-directional microphones without needing to collect real RIRs. By relying on specific angular regions and multiple room simulations, the model utilizes consistent time difference of arrival (TDOA) cues, or what we call delay contrast, to separate target and interference sources while remaining robust in various reverberation environments. We demonstrate the model is not only generalizable to a commercially available device with a slightly different microphone geometry, but also outperforms our previous work which uses one additional microphone on the same device. The model runs in real-time on-device and is suitable for low-latency streaming applications such as telephony and video conferencing.

Aswin Sivaraman, Scott Wisdom, Hakan Erdogan et al.

The recently-proposed mixture invariant training (MixIT) is an unsupervised method for training single-channel sound separation models in the sense that it does not require ground-truth isolated reference sources. In this paper, we investigate using MixIT to adapt a separation model on real far-field overlapping reverberant and noisy speech data from the AMI Corpus. The models are tested on real AMI recordings containing overlapping speech, and are evaluated subjectively by human listeners. To objectively evaluate our models, we also devise a synthetic AMI test set. For human evaluations on real recordings, we also propose a modification of the standard MUSHRA protocol to handle imperfect reference signals, which we call MUSHIRA. Holding network architectures constant, we find that a fine-tuned semi-supervised model yields the largest SI-SNR improvement, PESQ scores, and human listening ratings across synthetic and real datasets, outperforming unadapted generalist models trained on orders of magnitude more data. Our results show that unsupervised learning through MixIT enables model adaptation on real-world unlabeled spontaneous speech recordings.

Yuma Koizumi, Shigeki Karita, Scott Wisdom et al.

Single-channel speech enhancement (SE) is an important task in speech processing. A widely used framework combines an analysis/synthesis filterbank with a mask prediction network, such as the Conv-TasNet architecture. In such systems, the denoising performance and computational efficiency are mainly affected by the structure of the mask prediction network. In this study, we aim to improve the sequential modeling ability of Conv-TasNet architectures by integrating Conformer layers into a new mask prediction network. To make the model computationally feasible, we extend the Conformer using linear complexity attention and stacked 1-D dilated depthwise convolution layers. We trained the model on 3,396 hours of noisy speech data, and show that (i) the use of linear complexity attention avoids high computational complexity, and (ii) our model achieves higher scale-invariant signal-to-noise ratio than the improved time-dilated convolution network (TDCN++), an extended version of Conv-TasNet.

Scott Wisdom, Aren Jansen, Ron J. Weiss et al.

Supervised neural network training has led to significant progress on single-channel sound separation. This approach relies on ground truth isolated sources, which precludes scaling to widely available mixture data and limits progress on open-domain tasks. The recent mixture invariant training (MixIT) method enables training on in-the-wild data; however, it suffers from two outstanding problems. First, it produces models which tend to over-separate, producing more output sources than are present in the input. Second, the exponential computational complexity of the MixIT loss limits the number of feasible output sources. In this paper we address both issues. To combat over-separation we introduce new losses: sparsity losses that favor fewer output sources and a covariance loss that discourages correlated outputs. We also experiment with a semantic classification loss by predicting weak class labels for each mixture. To handle larger numbers of sources, we introduce an efficient approximation using a fast least-squares solution, projected onto the MixIT constraint set. Our experiments show that the proposed losses curtail over-separation and improve overall performance. The best performance is achieved using larger numbers of output sources, enabled by our efficient MixIT loss, combined with sparsity losses to prevent over-separation. On the FUSS test set, we achieve over 13 dB in multi-source SI-SNR improvement, while boosting single-source reconstruction SI-SNR by over 17 dB.

Soumi Maiti, Hakan Erdogan, Kevin Wilson et al.

We present an end-to-end deep network model that performs meeting diarization from single-channel audio recordings. End-to-end diarization models have the advantage of handling speaker overlap and enabling straightforward handling of discriminative training, unlike traditional clustering-based diarization methods. The proposed system is designed to handle meetings with unknown numbers of speakers, using variable-number permutation-invariant cross-entropy based loss functions. We introduce several components that appear to help with diarization performance, including a local convolutional network followed by a global self-attention module, multi-task transfer learning using a speaker identification component, and a sequential approach where the model is refined with a second stage. These are trained and validated on simulated meeting data based on LibriSpeech and LibriTTS datasets; final evaluations are done using LibriCSS, which consists of simulated meetings recorded using real acoustics via loudspeaker playback. The proposed model performs better than previously proposed end-to-end diarization models on these data.

Cong Han, Yi Luo, Chenda Li et al.

Leveraging additional speaker information to facilitate speech separation has received increasing attention in recent years. Recent research includes extracting target speech by using the target speaker's voice snippet and jointly separating all participating speakers by using a pool of additional speaker signals, which is known as speech separation using speaker inventory (SSUSI). However, all these systems ideally assume that the pre-enrolled speaker signals are available and are only evaluated on simple data configurations. In realistic multi-talker conversations, the speech signal contains a large proportion of non-overlapped regions, where we can derive robust speaker embedding of individual talkers. In this work, we adopt the SSUSI model in long recordings and propose a self-informed, clustering-based inventory forming scheme for long recording, where the speaker inventory is fully built from the input signal without the need for external speaker signals. Experiment results on simulated noisy reverberant long recording datasets show that the proposed method can significantly improve the separation performance across various conditions.

Desh Raj, Pavel Denisov, Zhuo Chen et al.

Multi-speaker speech recognition of unsegmented recordings has diverse applications such as meeting transcription and automatic subtitle generation. With technical advances in systems dealing with speech separation, speaker diarization, and automatic speech recognition (ASR) in the last decade, it has become possible to build pipelines that achieve reasonable error rates on this task. In this paper, we propose an end-to-end modular system for the LibriCSS meeting data, which combines independently trained separation, diarization, and recognition components, in that order. We study the effect of different state-of-the-art methods at each stage of the pipeline, and report results using task-specific metrics like SDR and DER, as well as downstream WER. Experiments indicate that the problem of overlapping speech for diarization and ASR can be effectively mitigated with the presence of a well-trained separation module. Our best system achieves a speaker-attributed WER of 12.7%, which is close to that of a non-overlapping ASR.

Nicolas Turpault, Romain Serizel, Scott Wisdom et al.

We propose a benchmark of state-of-the-art sound event detection systems (SED). We designed synthetic evaluation sets to focus on specific sound event detection challenges. We analyze the performance of the submissions to DCASE 2021 task 4 depending on time related modifications (time position of an event and length of clips) and we study the impact of non-target sound events and reverberation. We show that the localization in time of sound events is still a problem for SED systems. We also show that reverberation and non-target sound events are severely degrading the performance of the SED systems. In the latter case, sound separation seems like a promising solution.

Nicolas Turpault, Scott Wisdom, Hakan Erdogan et al.

Performing sound event detection on real-world recordings often implies dealing with overlapping target sound events and non-target sounds, also referred to as interference or noise. Until now these problems were mainly tackled at the classifier level. We propose to use sound separation as a pre-processing for sound event detection. In this paper we start from a sound separation model trained on the Free Universal Sound Separation dataset and the DCASE 2020 task 4 sound event detection baseline. We explore different methods to combine separated sound sources and the original mixture within the sound event detection. Furthermore, we investigate the impact of adapting the sound separation model to the sound event detection data on both the sound separation and the sound event detection.

Scott Wisdom, Efthymios Tzinis, Hakan Erdogan et al.

In recent years, rapid progress has been made on the problem of single-channel sound separation using supervised training of deep neural networks. In such supervised approaches, a model is trained to predict the component sources from synthetic mixtures created by adding up isolated ground-truth sources. Reliance on this synthetic training data is problematic because good performance depends upon the degree of match between the training data and real-world audio, especially in terms of the acoustic conditions and distribution of sources. The acoustic properties can be challenging to accurately simulate, and the distribution of sound types may be hard to replicate. In this paper, we propose a completely unsupervised method, mixture invariant training (MixIT), that requires only single-channel acoustic mixtures. In MixIT, training examples are constructed by mixing together existing mixtures, and the model separates them into a variable number of latent sources, such that the separated sources can be remixed to approximate the original mixtures. We show that MixIT can achieve competitive performance compared to supervised methods on speech separation. Using MixIT in a semi-supervised learning setting enables unsupervised domain adaptation and learning from large amounts of real world data without ground-truth source waveforms. In particular, we significantly improve reverberant speech separation performance by incorporating reverberant mixtures, train a speech enhancement system from noisy mixtures, and improve universal sound separation by incorporating a large amount of in-the-wild data.

Zhong-Qiu Wang, Hakan Erdogan, Scott Wisdom et al.

This work introduces sequential neural beamforming, which alternates between neural network based spectral separation and beamforming based spatial separation. Our neural networks for separation use an advanced convolutional architecture trained with a novel stabilized signal-to-noise ratio loss function. For beamforming, we explore multiple ways of computing time-varying covariance matrices, including factorizing the spatial covariance into a time-varying amplitude component and a time-invariant spatial component, as well as using block-based techniques. In addition, we introduce a multi-frame beamforming method which improves the results significantly by adding contextual frames to the beamforming formulations. We extensively evaluate and analyze the effects of window size, block size, and multi-frame context size for these methods. Our best method utilizes a sequence of three neural separation and multi-frame time-invariant spatial beamforming stages, and demonstrates an average improvement of 2.75 dB in scale-invariant signal-to-noise ratio and 14.2% absolute reduction in a comparative speech recognition metric across four challenging reverberant speech enhancement and separation tasks. We also use our three-speaker separation model to separate real recordings in the LibriCSS evaluation set into non-overlapping tracks, and achieve a better word error rate as compared to a baseline mask based beamformer.

Ilya Kavalerov, Scott Wisdom, Hakan Erdogan et al.

Recent deep learning approaches have achieved impressive performance on speech enhancement and separation tasks. However, these approaches have not been investigated for separating mixtures of arbitrary sounds of different types, a task we refer to as universal sound separation, and it is unknown how performance on speech tasks carries over to non-speech tasks. To study this question, we develop a dataset of mixtures containing arbitrary sounds, and use it to investigate the space of mask-based separation architectures, varying both the overall network architecture and the framewise analysis-synthesis basis for signal transformations. These network architectures include convolutional long short-term memory networks and time-dilated convolution stacks inspired by the recent success of time-domain enhancement networks like ConvTasNet. For the latter architecture, we also propose novel modifications that further improve separation performance. In terms of the framewise analysis-synthesis basis, we explore both a short-time Fourier transform (STFT) and a learnable basis, as used in ConvTasNet. For both of these bases, we also examine the effect of window size. In particular, for STFTs, we find that longer windows (25-50 ms) work best for speech/non-speech separation, while shorter windows (2.5 ms) work best for arbitrary sounds. For learnable bases, shorter windows (2.5 ms) work best on all tasks. Surprisingly, for universal sound separation, STFTs outperform learnable bases. Our best methods produce an improvement in scale-invariant signal-to-distortion ratio of over 13 dB for speech/non-speech separation and close to 10 dB for universal sound separation.

Takuya Yoshioka, Zhuo Chen, Changliang Liu et al.

Speaker independent continuous speech separation (SI-CSS) is a task of converting a continuous audio stream, which may contain overlapping voices of unknown speakers, into a fixed number of continuous signals each of which contains no overlapping speech segment. A separated, or cleaned, version of each utterance is generated from one of SI-CSS's output channels nondeterministically without being split up and distributed to multiple channels. A typical application scenario is transcribing multi-party conversations, such as meetings, recorded with microphone arrays. The output signals can be simply sent to a speech recognition engine because they do not include speech overlaps. The previous SI-CSS method uses a neural network trained with permutation invariant training and a data-driven beamformer and thus requires much processing latency. This paper proposes a low-latency SI-CSS method whose performance is comparable to that of the previous method in a microphone array-based meeting transcription task.This is achieved (1) by using a new speech separation network architecture combined with a double buffering scheme and (2) by performing enhancement with a set of fixed beamformers followed by a neural post-filter.

Jonathan Le Roux, Scott Wisdom, Hakan Erdogan et al.

In speech enhancement and source separation, signal-to-noise ratio is a ubiquitous objective measure of denoising/separation quality. A decade ago, the BSS_eval toolkit was developed to give researchers worldwide a way to evaluate the quality of their algorithms in a simple, fair, and hopefully insightful way: it attempted to account for channel variations, and to not only evaluate the total distortion in the estimated signal but also split it in terms of various factors such as remaining interference, newly added artifacts, and channel errors. In recent years, hundreds of papers have been relying on this toolkit to evaluate their proposed methods and compare them to previous works, often arguing that differences on the order of 0.1 dB proved the effectiveness of a method over others. We argue here that the signal-to-distortion ratio (SDR) implemented in the BSS_eval toolkit has generally been improperly used and abused, especially in the case of single-channel separation, resulting in misleading results. We propose to use a slightly modified definition, resulting in a simpler, more robust measure, called scale-invariant SDR (SI-SDR). We present various examples of critical failure of the original SDR that SI-SDR overcomes.

Takuya Yoshioka, Hakan Erdogan, Zhuo Chen et al.

The goal of this work is to develop a meeting transcription system that can recognize speech even when utterances of different speakers are overlapped. While speech overlaps have been regarded as a major obstacle in accurately transcribing meetings, a traditional beamformer with a single output has been exclusively used because previously proposed speech separation techniques have critical constraints for application to real meetings. This paper proposes a new signal processing module, called an unmixing transducer, and describes its implementation using a windowed BLSTM. The unmixing transducer has a fixed number, say J, of output channels, where J may be different from the number of meeting attendees, and transforms an input multi-channel acoustic signal into J time-synchronous audio streams. Each utterance in the meeting is separated and emitted from one of the output channels. Then, each output signal can be simply fed to a speech recognition back-end for segmentation and transcription. Our meeting transcription system using the unmixing transducer outperforms a system based on a state-of-the-art neural mask-based beamformer by 10.8%. Significant improvements are observed in overlapped segments. To the best of our knowledge, this is the first report that applies overlapped speech recognition to unconstrained real meeting audio.

Zhong Meng, Shinji Watanabe, John R. Hershey et al.

Far-field speech recognition in noisy and reverberant conditions remains a challenging problem despite recent deep learning breakthroughs. This problem is commonly addressed by acquiring a speech signal from multiple microphones and performing beamforming over them. In this paper, we propose to use a recurrent neural network with long short-term memory (LSTM) architecture to adaptively estimate real-time beamforming filter coefficients to cope with non-stationary environmental noise and dynamic nature of source and microphones positions which results in a set of timevarying room impulse responses. The LSTM adaptive beamformer is jointly trained with a deep LSTM acoustic model to predict senone labels. Further, we use hidden units in the deep LSTM acoustic model to assist in predicting the beamforming filter coefficients. The proposed system achieves 7.97% absolute gain over baseline systems with no beamforming on CHiME-3 real evaluation set.

Ahmet E. Bulut, Hakan Demir, Yusuf Ziya Isik et al.

This paper investigates the application of the probabilistic linear discriminant analysis (PLDA) to speaker diarization of telephone conversations. We introduce using a variational Bayes (VB) approach for inference under a PLDA model for modeling segmental i-vectors in speaker diarization. Deterministic annealing (DA) algorithm is imposed in order to avoid local optimal solutions in VB iterations. We compare our proposed system with a well-known system that applies k-means clustering on principal component analysis (PCA) coefficients of segmental i-vectors. We used summed channel telephone data from the National Institute of Standards and Technology (NIST) 2008 Speaker Recognition Evaluation (SRE) as the test set in order to evaluate the performance of the proposed system. We achieve about 20% relative improvement in Diarization Error Rate (DER) compared to the baseline system.

Cagatay Karabat, Mehmet Sabir Kiraz, Hakan Erdogan et al.

In this paper, we propose a new biometric verification and template protection system which we call the THRIVE system. The system includes novel enrollment and authentication protocols based on threshold homomorphic cryptosystem where the private key is shared between a user and the verifier. In the THRIVE system, only encrypted binary biometric templates are stored in the database and verification is performed via homomorphically randomized templates, thus, original templates are never revealed during the authentication stage. The THRIVE system is designed for the malicious model where the cheating party may arbitrarily deviate from the protocol specification. Since threshold homomorphic encryption scheme is used, a malicious database owner cannot perform decryption on encrypted templates of the users in the database. Therefore, security of the THRIVE system is enhanced using a two-factor authentication scheme involving the user's private key and the biometric data. We prove security and privacy preservation capability of the proposed system in the simulation-based model with no assumption. The proposed system is suitable for applications where the user does not want to reveal her biometrics to the verifier in plain form but she needs to proof her physical presence by using biometrics. The system can be used with any biometric modality and biometric feature extraction scheme whose output templates can be binarized. The overall connection time for the proposed THRIVE system is estimated to be 336 ms on average for 256-bit biohash vectors on a desktop PC running with quad-core 3.2 GHz CPUs at 10 Mbit/s up/down link connection speed. Consequently, the proposed system can be efficiently used in real life applications.

Emad M. Grais, Mehmet Umut Sen, Hakan Erdogan

In this paper, a novel approach for single channel source separation (SCSS) using a deep neural network (DNN) architecture is introduced. Unlike previous studies in which DNN and other classifiers were used for classifying time-frequency bins to obtain hard masks for each source, we use the DNN to classify estimated source spectra to check for their validity during separation. In the training stage, the training data for the source signals are used to train a DNN. In the separation stage, the trained DNN is utilized to aid in estimation of each source in the mixed signal. Single channel source separation problem is formulated as an energy minimization problem where each source spectra estimate is encouraged to fit the trained DNN model and the mixed signal spectrum is encouraged to be written as a weighted sum of the estimated source spectra. The proposed approach works regardless of the energy scale differences between the source signals in the training and separation stages. Nonnegative matrix factorization (NMF) is used to initialize the DNN estimate for each source. The experimental results show that using DNN initialized by NMF for source separation improves the quality of the separated signal compared with using NMF for source separation.

Emad M. Grais, Hakan Erdogan

We propose a new method to enforce priors on the solution of the nonnegative matrix factorization (NMF). The proposed algorithm can be used for denoising or single-channel source separation (SCSS) applications. The NMF solution is guided to follow the Minimum Mean Square Error (MMSE) estimates under Gaussian mixture prior models (GMM) for the source signal. In SCSS applications, the spectra of the observed mixed signal are decomposed as a weighted linear combination of trained basis vectors for each source using NMF. In this work, the NMF decomposition weight matrices are treated as a distorted image by a distortion operator, which is learned directly from the observed signals. The MMSE estimate of the weights matrix under GMM prior and log-normal distribution for the distortion is then found to improve the NMF decomposition results. The MMSE estimate is embedded within the optimization objective to form a novel regularized NMF cost function. The corresponding update rules for the new objectives are derived in this paper. Experimental results show that, the proposed regularized NMF algorithm improves the source separation performance compared with using NMF without prior or with other prior models.