Saurabh Kataria

Saurabh Kataria, Xiao Hu

Audio-Language Models (ALMs) are making strides in understanding speech and non-speech audio. However, domain-specialist Foundation Models (FMs) remain the best for closed-ended speech processing tasks such as Speech Emotion Recognition (SER). Using ALMs for Zero-shot SER is a popular choice, but their potential to work with specialists to achieve state-of-the-art (SOTA) performance remains unexplored. We propose ZS-Fuse, a late-fusion method that combines zero-shot emotion estimates from a dual-encoder ALM with specialist FMs. To handle ambiguity in emotions and sensitivity to prompt choice, 1) we use a simple prompt ensemble and 2) suggest a novel technique called prompt amplification, which repeats audio and text queries to discover stronger zero-shot capabilities. We demonstrate the efficacy of our technique by evaluating ZS-Fuse with three dual-encoder ALMs and two FMs, and report improvements over SOTA baselines, such as WavLM-Large, on three speech emotion recognition datasets.

Zhaoliang Chen, Cheng Ding, Saurabh Kataria et al.

This study introduces a novel application of a Generative Pre-trained Transformer (GPT) model tailored for photoplethysmography (PPG) signals, serving as a foundation model for various downstream tasks. Adapting the standard GPT architecture to suit the continuous characteristics of PPG signals, our approach demonstrates promising results. Our models are pre-trained on our extensive dataset that contains more than 200 million 30s PPG samples. We explored different supervised fine-tuning techniques to adapt our model to downstream tasks, resulting in performance comparable to or surpassing current state-of-the-art (SOTA) methods in tasks like atrial fibrillation detection. A standout feature of our GPT model is its inherent capability to perform generative tasks such as signal denoising effectively, without the need for further fine-tuning. This success is attributed to the generative nature of the GPT framework.

Juntong Ni, Saurabh Kataria, Shengpu Tang et al.

Photoplethysmography (PPG) is widely used in wearable health monitoring, yet large PPG foundation models remain difficult to deploy on resource-limited devices. We present PPG-Distill, a knowledge distillation framework that transfers both global and local knowledge through prediction-, feature-, and patch-level distillation. PPG-Distill incorporates morphology distillation to preserve local waveform patterns and rhythm distillation to capture inter-patch temporal structures. On heart rate estimation and atrial fibrillation detection, PPG-Distill improves student performance by up to 21.8% while achieving 7X faster inference and reducing memory usage by 19X, enabling efficient PPG analysis on wearables.

Sudarshan Srinivasa Ramanujam, Antonio Alonso, Saurabh Kataria et al.

In large scale recommendation systems like the LinkedIn Feed, the retrieval stage is critical for narrowing hundreds of millions of potential candidates to a manageable subset for ranking. LinkedIn's Feed serves suggested content from outside of the member's network (based on the member's topical interests), where 2000 candidates are retrieved from a pool of hundreds of millions candidate with a latency budget of a few milliseconds and inbound QPS of several thousand per second. This paper presents a novel retrieval approach that fine-tunes a large causal language model (Meta's LLaMA 3) as a dual encoder to generate high quality embeddings for both users (members) and content (items), using only textual input. We describe the end to end pipeline, including prompt design for embedding generation, techniques for fine-tuning at LinkedIn's scale, and infrastructure for low latency, cost effective online serving. We share our findings on how quantizing numerical features in the prompt enables the information to get properly encoded in the embedding, facilitating greater alignment between the retrieval and ranking layer. The system was evaluated using offline metrics and an online A/B test, which showed substantial improvements in member engagement. We observed significant gains among newer members, who often lack strong network connections, indicating that high-quality suggested content aids retention. This work demonstrates how generative language models can be effectively adapted for real time, high throughput retrieval in industrial applications.

Saurabh Kataria, Jesús Villalba, Najim Dehak

Deep learning based speech denoising still suffers from the challenge of improving perceptual quality of enhanced signals. We introduce a generalized framework called Perceptual Ensemble Regularization Loss (PERL) built on the idea of perceptual losses. Perceptual loss discourages distortion to certain speech properties and we analyze it using six large-scale pre-trained models: speaker classification, acoustic model, speaker embedding, emotion classification, and two self-supervised speech encoders (PASE+, wav2vec 2.0). We first build a strong baseline (w/o PERL) using Conformer Transformer Networks on the popular enhancement benchmark called VCTK-DEMAND. Using auxiliary models one at a time, we find acoustic event and self-supervised model PASE+ to be most effective. Our best model (PERL-AE) only uses acoustic event model (utilizing AudioSet) to outperform state-of-the-art methods on major perceptual metrics. To explore if denoising can leverage full framework, we use all networks but find that our seven-loss formulation suffers from the challenges of Multi-Task Learning. Finally, we report a critical observation that state-of-the-art Multi-Task weight learning methods cannot outperform hand tuning, perhaps due to challenges of domain mismatch and weak complementarity of losses.

Phani Sankar Nidadavolu, Saurabh Kataria, Paola García-Perera et al.

We investigated an enhancement and a domain adaptation approach to make speaker verification systems robust to perturbations of far-field speech. In the enhancement approach, using paired (parallel) reverberant-clean speech, we trained a supervised Generative Adversarial Network (GAN) along with a feature mapping loss. For the domain adaptation approach, we trained a Cycle Consistent Generative Adversarial Network (CycleGAN), which maps features from far-field domain to the speaker embedding training domain. This was trained on unpaired data in an unsupervised manner. Both networks, termed Supervised Enhancement Network (SEN) and Domain Adaptation Network (DAN) respectively, were trained with multi-task objectives in (filter-bank) feature domain. On a simulated test setup, we first note the benefit of using feature mapping (FM) loss along with adversarial loss in SEN. Then, we tested both supervised and unsupervised approaches on several real noisy datasets. We observed relative improvements ranging from 2% to 31% in terms of DCF. Using three training schemes, we also establish the effectiveness of the novel DAN approach.

Saurabh Kataria, Phani Sankar Nidadavolu, Jesús Villalba et al.

Data augmentation is conventionally used to inject robustness in Speaker Verification systems. Several recently organized challenges focus on handling novel acoustic environments. Deep learning based speech enhancement is a modern solution for this. Recently, a study proposed to optimize the enhancement network in the activation space of a pre-trained auxiliary network. This methodology, called deep feature loss, greatly improved over the state-of-the-art conventional x-vector based system on a children speech dataset called BabyTrain. This work analyzes various facets of that approach and asks few novel questions in that context. We first search for optimal number of auxiliary network activations, training data, and enhancement feature dimension. Experiments reveal the importance of Signal-to-Noise Ratio filtering that we employ to create a large, clean, and naturalistic corpus for enhancement network training. To counter the "mismatch" problem in enhancement, we find enhancing front-end (x-vector network) data helpful while harmful for the back-end (Probabilistic Linear Discriminant Analysis (PLDA)). Importantly, we find enhanced signals contain complementary information to original. Established by combining them in front-end, this gives ~40% relative improvement over the baseline. We also do an ablation study to remove a noise class from x-vector data augmentation and, for such systems, we establish the utility of enhancement regardless of whether it has seen that noise class itself during training. Finally, we design several dereverberation schemes to conclude ineffectiveness of deep feature loss enhancement scheme for this task.

Paola Garcia, Jesus Villalba, Herve Bredin et al.

This paper presents the problems and solutions addressed at the JSALT workshop when using a single microphone for speaker detection in adverse scenarios. The main focus was to tackle a wide range of conditions that go from meetings to wild speech. We describe the research threads we explored and a set of modules that was successful for these scenarios. The ultimate goal was to explore speaker detection; but our first finding was that an effective diarization improves detection, and not having a diarization stage impoverishes the performance. All the different configurations of our research agree on this fact and follow a main backbone that includes diarization as a previous stage. With this backbone, we analyzed the following problems: voice activity detection, how to deal with noisy signals, domain mismatch, how to improve the clustering; and the overall impact of previous stages in the final speaker detection. In this paper, we show partial results for speaker diarizarion to have a better understanding of the problem and we present the final results for speaker detection.

Phani Sankar Nidadavolu, Saurabh Kataria, Jesús Villalba et al.

The task of making speaker verification systems robust to adverse scenarios remain a challenging and an active area of research. We developed an unsupervised feature enhancement approach in log-filter bank domain with the end goal of improving speaker verification performance. We experimented with using both real speech recorded in adverse environments and degraded speech obtained by simulation to train the enhancement systems. The effectiveness of the approach was shown by testing on several real, simulated noisy, and reverberant test sets. The approach yielded significant improvements on both real and simulated sets when data augmentation was not used in speaker verification pipeline or augmentation was used only during x-vector training. When data augmentation was used for x-vector and PLDA training, our enhancement approach yielded slight improvements.

Phani Sankar Nidadavolu, Saurabh Kataria, Jesús Villalba et al.

Current speaker recognition technology provides great performance with the x-vector approach. However, performance decreases when the evaluation domain is different from the training domain, an issue usually addressed with domain adaptation approaches. Recently, unsupervised domain adaptation using cycle-consistent Generative Adversarial Netorks (CycleGAN) has received a lot of attention. CycleGAN learn mappings between features of two domains given non-parallel data. We investigate their effectiveness in low resource scenario i.e. when limited amount of target domain data is available for adaptation, a case unexplored in previous works. We experiment with two adaptation tasks: microphone to telephone and a novel reverberant to clean adaptation with the end goal of improving speaker recognition performance. Number of speakers present in source and target domains are 7000 and 191 respectively. By adding noise to the target domain during CycleGAN training, we were able to achieve better performance compared to the adaptation system whose CycleGAN was trained on a larger target data. On reverberant to clean adaptation task, our models improved EER by 18.3% relative on VOiCES dataset compared to a system trained on clean data. They also slightly improved over the state-of-the-art Weighted Prediction Error (WPE) de-reverberation algorithm.

Saurabh Kataria, Phani Sankar Nidadavolu, Jesús Villalba et al.

Speaker Verification still suffers from the challenge of generalization to novel adverse environments. We leverage on the recent advancements made by deep learning based speech enhancement and propose a feature-domain supervised denoising based solution. We propose to use Deep Feature Loss which optimizes the enhancement network in the hidden activation space of a pre-trained auxiliary speaker embedding network. We experimentally verify the approach on simulated and real data. A simulated testing setup is created using various noise types at different SNR levels. For evaluation on real data, we choose BabyTrain corpus which consists of children recordings in uncontrolled environments. We observe consistent gains in every condition over the state-of-the-art augmented Factorized-TDNN x-vector system. On BabyTrain corpus, we observe relative gains of 10.38% and 12.40% in minDCF and EER respectively.

Fenglong Ma, Radha Chitta, Saurabh Kataria et al.

Question answering is an important and difficult task in the natural language processing domain, because many basic natural language processing tasks can be cast into a question answering task. Several deep neural network architectures have been developed recently, which employ memory and inference components to memorize and reason over text information, and generate answers to questions. However, a major drawback of many such models is that they are capable of only generating single-word answers. In addition, they require large amount of training data to generate accurate answers. In this paper, we introduce the Long-Term Memory Network (LTMN), which incorporates both an external memory module and a Long Short-Term Memory (LSTM) module to comprehend the input data and generate multi-word answers. The LTMN model can be trained end-to-end using back-propagation and requires minimal supervision. We test our model on two synthetic data sets (based on Facebook's bAbI data set) and the real-world Stanford question answering data set, and show that it can achieve state-of-the-art performance.

Clément Gaultier, Saurabh Kataria, Antoine Deleforge

This paper introduces a new paradigm for sound source lo-calization referred to as virtual acoustic space traveling (VAST) and presents a first dataset designed for this purpose. Existing sound source localization methods are either based on an approximate physical model (physics-driven) or on a specific-purpose calibration set (data-driven). With VAST, the idea is to learn a mapping from audio features to desired audio properties using a massive dataset of simulated room impulse responses. This virtual dataset is designed to be maximally representative of the potential audio scenes that the considered system may be evolving in, while remaining reasonably compact. We show that virtually-learned mappings on this dataset generalize to real data, overcoming some intrinsic limitations of traditional binaural sound localization methods based on time differences of arrival.

Saurabh Kataria, Clément Gaultier, Antoine Deleforge

This paper introduces a new framework for supervised sound source localization referred to as virtually-supervised learning. An acoustic shoe-box room simulator is used to generate a large number of binaural single-source audio scenes. These scenes are used to build a dataset of spatial binaural features annotated with acoustic properties such as the 3D source position and the walls' absorption coefficients. A probabilistic high- to low-dimensional regression framework is used to learn a mapping from these features to the acoustic properties. Results indicate that this mapping successfully estimates the azimuth and elevation of new sources, but also their range and even the walls' absorption coefficients solely based on binaural signals. Results also reveal that incorporating random-diffusion effects in the data significantly improves the estimation of all parameters.

Saurabh Kataria

We consider the problem of learning distributed representations for tags from their associated content for the task of tag recommendation. Considering tagging information is usually very sparse, effective learning from content and tag association is very crucial and challenging task. Recently, various neural representation learning models such as WSABIE and its variants show promising performance, mainly due to compact feature representations learned in a semantic space. However, their capacity is limited by a linear compositional approach for representing tags as sum of equal parts and hurt their performance. In this work, we propose a neural feedback relevance model for learning tag representations with weighted feature representations. Our experiments on two widely used datasets show significant improvement for quality of recommendations over various baselines.

Arvind Agarwal, Saurabh Kataria

In this paper, we present a learning method for sequence labeling tasks in which each example sequence has multiple label sequences. Our method learns multiple models, one model for each label sequence. Each model computes the joint probability of all label sequences given the example sequence. Although each model considers all label sequences, its primary focus is only one label sequence, and therefore, each model becomes a task-specific model, for the task belonging to that primary label. Such multiple models are learned {\it simultaneously} by facilitating the learning transfer among models through {\it explicit parameter sharing}. We experiment the proposed method on two applications and show that our method significantly outperforms the state-of-the-art method.