Shlomo E. Chazan

Aviad Eisenberg, Sharon Gannot, Shlomo E. Chazan

In this paper we present a unified time-frequency method for speaker extraction in clean and noisy conditions. Given a mixed signal, along with a reference signal, the common approaches for extracting the desired speaker are either applied in the time-domain or in the frequency-domain. In our approach, we propose a Siamese-Unet architecture that uses both representations. The Siamese encoders are applied in the frequency-domain to infer the embedding of the noisy and reference spectra, respectively. The concatenated representations are then fed into the decoder to estimate the real and imaginary components of the desired speaker, which are then inverse-transformed to the time-domain. The model is trained with the Scale-Invariant Signal-to-Distortion Ratio (SI-SDR) loss to exploit the time-domain information. The time-domain loss is also regularized with frequency-domain loss to preserve the speech patterns. Experimental results demonstrate that the unified approach is not only very easy to train, but also provides superior results as compared with state-of-the-art (SOTA) Blind Source Separation (BSS) methods, as well as commonly used speaker extraction approach.

Tomer Wullach, Shlomo E. Chazan

Automatic Speech Recognition (ASR) systems frequently use a search-based decoding strategy aiming to find the best attainable transcript by considering multiple candidates. One prominent speech recognition decoding heuristic is beam search, which seeks the transcript with the greatest likelihood computed using the predicted distribution. While showing substantial performance gains in various tasks, beam search loses some of its effectiveness when the predicted probabilities are highly confident, i.e., the predicted distribution is massed for a single or very few classes. We show that recently proposed Self-Supervised Learning (SSL)-based ASR models tend to yield exceptionally confident predictions that may hamper beam search from truly considering a diverse set of candidates. We perform a layer analysis to reveal and visualize how predictions evolve, and propose a decoding procedure that improves the performance of fine-tuned ASR models. Our proposed approach does not require further training beyond the original fine-tuning, nor additional model parameters. In fact, we find that our proposed method requires significantly less inference computation than current approaches. We propose aggregating the top M layers, potentially leveraging useful information encoded in intermediate layers, and relaxing model confidence. We demonstrate the effectiveness of our approach by conducting an empirical study on varying amounts of labeled resources and different model sizes, showing consistent improvements in particular when applied to low-resource scenarios.

Tomer Wullach, Shlomo E. Chazan

Recently proposed speech recognition systems are designed to predict using representations generated by their top layers, employing greedy decoding which isolates each timestep from the rest of the sequence. Aiming for improved performance, a beam search algorithm is frequently utilized and a language model is incorporated to assist with ranking the top candidates. In this work, we experiment with several speech recognition models and find that logits predicted using the top layers may hamper beam search from achieving optimal results. Specifically, we show that fined-tuned Wav2Vec 2.0 and HuBERT yield highly confident predictions, and hypothesize that the predictions are based on local information and may not take full advantage of the information encoded in intermediate layers. To this end, we perform a layer analysis to reveal and visualize how predictions evolve throughout the inference flow. We then propose a prediction method that aggregates the top M layers, potentially leveraging useful information encoded in intermediate layers and relaxing model confidence. We showcase the effectiveness of our approach via beam search decoding, conducting our experiments on Librispeech test and dev sets and achieving WER, and CER reduction of up to 10% and 22%, respectively.

Tomer Wullach, Shlomo E. Chazan

The challenges facing speech recognition systems, such as variations in pronunciations, adverse audio conditions, and the scarcity of labeled data, emphasize the necessity for a post-processing step that corrects recurring errors. Previous research has shown the advantages of employing dedicated error correction models, yet training such models requires large amounts of labeled data which is not easily obtained. To overcome this limitation, synthetic transcribed-like data is often utilized, however, bridging the distribution gap between transcribed errors and synthetic noise is not trivial. In this paper, we demonstrate that the performance of correction models can be significantly increased by training solely using synthetic data. Specifically, we empirically show that: (1) synthetic data generated using the error distribution derived from a set of transcribed data outperforms the common approach of applying random perturbations; (2) applying language-specific adjustments to the vocabulary of a BPE tokenizer strike a balance between adapting to unseen distributions and retaining knowledge of transcribed errors. We showcase the benefits of these key observations, and evaluate our approach using multiple languages, speech recognition systems and prominent speech recognition datasets.

Aviad Eisenberg, Sharon Gannot, Shlomo E. Chazan

This paper introduces a multi-microphone method for extracting a desired speaker from a mixture involving multiple speakers and directional noise in a reverberant environment. In this work, we propose leveraging the instantaneous relative transfer function (RTF), estimated from a reference utterance recorded in the same position as the desired source. The effectiveness of the RTF-based spatial cue is compared with direction of arrival (DOA)-based spatial cue and the conventional spectral embedding. Experimental results in challenging acoustic scenarios demonstrate that using spatial cues yields better performance than the spectral-based cue and that the instantaneous RTF outperforms the DOA-based spatial cue.

Ori Shapira, Shlomo E. Chazan, Amir DN Cohen · amazon-science

With the increasing prevalence of recorded human speech, spoken language understanding (SLU) is essential for its efficient processing. In order to process the speech, it is commonly transcribed using automatic speech recognition technology. This speech-to-text transition introduces errors into the transcripts, which subsequently propagate to downstream NLP tasks, such as dialogue summarization. While it is known that transcript noise affects downstream tasks, a systematic approach to analyzing its effects across different noise severities and types has not been addressed. We propose a configurable framework for assessing task models in diverse noisy settings, and for examining the impact of transcript-cleaning techniques. The framework facilitates the investigation of task model behavior, which can in turn support the development of effective SLU solutions. We exemplify the utility of our framework on three SLU tasks and four task models, offering insights regarding the effect of transcript noise on tasks in general and models in particular. For instance, we find that task models can tolerate a certain level of noise, and are affected differently by the types of errors in the transcript.

Shlomo E. Chazan, Jacob Goldberger, Sharon Gannot

In this study we present a mixture of deep experts (MoDE) neural-network architecture for single microphone speech enhancement. Our architecture comprises a set of deep neural networks (DNNs), each of which is an 'expert' in a different speech spectral pattern such as phoneme. A gating DNN is responsible for the latent variables which are the weights assigned to each expert's output given a speech segment. The experts estimate a mask from the noisy input and the final mask is then obtained as a weighted average of the experts' estimates, with the weights determined by the gating DNN. A soft spectral attenuation, based on the estimated mask, is then applied to enhance the noisy speech signal. As a byproduct, we gain reduction at the complexity in test time. We show that the experts specialization allows better robustness to unfamiliar noise types.

Shlomo E. Chazan, Lior Wolf, Eliya Nachmani et al.

We present a unified network for voice separation of an unknown number of speakers. The proposed approach is composed of several separation heads optimized together with a speaker classification branch. The separation is carried out in the time domain, together with parameter sharing between all separation heads. The classification branch estimates the number of speakers while each head is specialized in separating a different number of speakers. We evaluate the proposed model under both clean and noisy reverberant set-tings. Results suggest that the proposed approach is superior to the baseline model by a significant margin. Additionally, we present a new noisy and reverberant dataset of up to five different speakers speaking simultaneously.

Hodaya Hammer, Shlomo E. Chazan, Jacob Goldberger et al.

In this paper, we present a deep neural network-based online multi-speaker localisation algorithm. Following the W-disjoint orthogonality principle in the spectral domain, each time-frequency (TF) bin is dominated by a single speaker, and hence by a single direction of arrival (DOA). A fully convolutional network is trained with instantaneous spatial features to estimate the DOA for each TF bin. The high resolution classification enables the network to accurately and simultaneously localize and track multiple speakers, both static and dynamic. Elaborated experimental study using both simulated and real-life recordings in static and dynamic scenarios, confirms that the proposed algorithm outperforms both classic and recent deep-learning-based algorithms.

Shlomo E. Chazan, Sharon Gannot, Jacob Goldberger

In this paper we propose a Deep Autoencoder MIxture Clustering (DAMIC) algorithm based on a mixture of deep autoencoders where each cluster is represented by an autoencoder. A clustering network transforms the data into another space and then selects one of the clusters. Next, the autoencoder associated with this cluster is used to reconstruct the data-point. The clustering algorithm jointly learns the nonlinear data representation and the set of autoencoders. The optimal clustering is found by minimizing the reconstruction loss of the mixture of autoencoder network. Unlike other deep clustering algorithms, no regularization term is needed to avoid data collapsing to a single point. Our experimental evaluations on image and text corpora show significant improvement over state-of-the-art methods.

Ori Ernst, Shlomo E. Chazan, Sharon Gannot et al.

Speech derverberation using a single microphone is addressed in this paper. Motivated by the recent success of the fully convolutional networks (FCN) in many image processing applications, we investigate their applicability to enhance the speech signal represented by short-time Fourier transform (STFT) images. We present two variations: a "U-Net" which is an encoder-decoder network with skip connections and a generative adversarial network (GAN) with U-Net as generator, which yields a more intuitive cost function for training. To evaluate our method we used the data from the REVERB challenge, and compared our results to other methods under the same conditions. We have found that our method outperforms the competing methods in most cases.

Shlomo E. Chazan, Jacob Goldberger, Sharon Gannot

In this study we present a Deep Mixture of Experts (DMoE) neural-network architecture for single microphone speech enhancement. By contrast to most speech enhancement algorithms that overlook the speech variability mainly caused by phoneme structure, our framework comprises a set of deep neural networks (DNNs), each one of which is an 'expert' in enhancing a given speech type corresponding to a phoneme. A gating DNN determines which expert is assigned to a given speech segment. A speech presence probability (SPP) is then obtained as a weighted average of the expert SPP decisions, with the weights determined by the gating DNN. A soft spectral attenuation, based on the SPP, is then applied to enhance the noisy speech signal. The experts and the gating components of the DMoE network are trained jointly. As part of the training, speech clustering into different subsets is performed in an unsupervised manner. Therefore, unlike previous methods, a phoneme-labeled database is not required for the training procedure. A series of experiments with different noise types verified the applicability of the new algorithm to the task of speech enhancement. The proposed scheme outperforms other schemes that either do not consider phoneme structure or use a simpler training methodology.

Shlomo E. Chazan, Jacob Goldberger, Sharon Gannot

In this paper we present a single-microphone speech enhancement algorithm. A hybrid approach is proposed merging the generative mixture of Gaussians (MoG) model and the discriminative neural network (NN). The proposed algorithm is executed in two phases, the training phase, which does not recur, and the test phase. First, the noise-free speech power spectral density (PSD) is modeled as a MoG, representing the phoneme based diversity in the speech signal. An NN is then trained with phoneme labeled database for phoneme classification with mel-frequency cepstral coefficients (MFCC) as the input features. Given the phoneme classification results, a speech presence probability (SPP) is obtained using both the generative and discriminative models. Soft spectral subtraction is then executed while simultaneously, the noise estimation is updated. The discriminative NN maintain the continuity of the speech and the generative phoneme-based MoG preserves the speech spectral structure. Extensive experimental study using real speech and noise signals is provided. We also compare the proposed algorithm with alternative speech enhancement algorithms. We show that we obtain a significant improvement over previous methods in terms of both speech quality measures and speech recognition results.