Text-to-speech (TTS) systems are designed to synthesize natural and expressive speech, adapt to an unseen voice, and capture the speaking style of an unseen speaker by converting text into speech. The introduction of an unseen speaker’s speaking style into a TTS system offers a wide range of application scenarios, including personal assistant, news broadcast, and audio navigation, among others. The style of the speech varies from person to person and every person exhibits his or her style of speaking that is determined by the language, demography, culture and other factors. Style is best captured by the prosody of a signal. It is an ongoing research area with numerous real-world applications that produces high-quality multi-speaker voice synthesis while taking into account prosody and in a zero-shot manner. Despite the fact that several efforts have been made in this area, it continues to be an interesting and difficult topic to solve. In this paper, we present a novel zero-shot multi-speaker speech synthesis approach (ZSM-SS) that leverages the normalization architecture and speaker encoder with non-autoregressive multi-head attention driven encoder-decoder architecture. Given an input text and a reference speech sample of an unseen person, ZSM-SS can generate speech in that person’s style in a zero-shot manner. Additionally, we demonstrate how the affine parameters of normalization help in capturing the prosodic features such as energy and fundamental frequency in a disentangled fashion and can be used to generate morphed speech output. We generate the 256 dimensional speaker embedding using a speaker encoder based on wav2vec2.0 based architecture. We demonstrate the efficacy of our proposed architecture on multi-speaker VCTK [1] and LibriTTS [2] datasets, using visualization of hessian of proposed model, multiple quantitative metrics that measure generated speech distortion and MOS, along with speaker embedding analysis of the proposed speaker encoder model.

Stickers are popularly used in messaging apps such as Hike to visually express a nuanced range of thoughts and utterances to convey exaggerated emotions. However, discovering the right sticker from a large and ever expanding pool of stickers while chatting can be cumbersome. In this paper, we describe a system for recommending stickers in real time as the user is typing based on the context of the conversation. We decompose the sticker recommendation (SR) problem into two steps. First, we predict the message that the user is likely to send in the chat. Second, we substitute the predicted message with an appropriate sticker. Majority of Hike's messages are in the form of text which is transliterated from users' native language to the Roman script. This leads to numerous orthographic variations of the same message and makes accurate message prediction challenging. To address this issue, we learn dense representations of chat messages employing character level convolution network in an unsupervised manner. We use them to cluster the messages that have the same meaning. In the subsequent steps, we predict the message cluster instead of the message. Our approach does not depend on human labelled data (except for validation), leading to fully automatic updation and tuning pipeline for the underlying models. We also propose a novel hybrid message prediction model, which can run with low latency on low-end phones that have severe computational limitations. Our described system has been deployed for more than 6 months and is being used by millions of users along with hundreds of thousands of expressive stickers.

With increasing popularity of Netflix, Yahoo! Video, etc., interactive multimedia services such as video-on-demand (VoD) provide an interesting and rich field of research. The advent of smarter wireless devices has surged the need for such services through wireless connectivity. However, personalization of individual user needs, reducing latency, coupled with maintaining low operational costs provides a challenging problem. In this paper, we propose an efficient VoD system, for wireless mobile devices, based on a novel caching algorithm, Intelligent Network Caching Algorithm (INCA) using analytics-driven look ahead scheme for both prefetch and replacement policies to deliver higher performance. This enables enhanced Quality Of Experience (QoE) of users with limited infrastructural changes and low operational cost. Alongside, we develop theoretical formulation of the QoE optimization problem that lies at the intersection of MPC (Markov Predictive Control) and MDP (Markov Decision Process). Empirical analysis over realistic user video query logs demonstrate better cache hit rate and QoE with low prefetch bandwidth, compared to existing caching schemes.

The unparalleled growth and popularity of the Internet coupled with the advent of diverse modern applications such as search engines, on-line transactions, climate warning systems, etc., has catered to an unprecedented expanse in the volume of data stored world-wide. Efficient storage, management, and processing of such massively exponential amount of data has emerged as a central theme of research in this direction. Detection and removal of redundancies and duplicates in real-time from such multi-trillion record-set to bolster resource and compute efficiency constitutes a challenging area of study. The infeasibility of storing the entire data from potentially unbounded data streams, with the need for precise elimination of duplicates calls for intelligent approximate duplicate detection algorithms. The literature hosts numerous works based on the well-known probabilistic bitmap structure, Bloom Filter and its variants. In this paper we propose a novel data structure, Streaming Quotient Filter, (SQF) for efficient detection and removal of duplicates in data streams. SQF intelligently stores the signatures of elements arriving on a data stream, and along with an eviction policy provides near zero false positive and false negative rates. We show that the near optimal performance of SQF is achieved with a very low memory requirement, making it ideal for real-time memory-efficient de-duplication applications having an extremely low false positive and false negative tolerance rates. We present detailed theoretical analysis of the working of SQF, providing a guarantee on its performance. Empirically, we compare SQF to alternate methods and show that the proposed method is superior in terms of memory and accuracy compared to the existing solutions. We also discuss Dynamic SQF for evolving streams and the parallel implementation of SQF.

With the explosion of information stored world-wide, data intensive computing has emerged as a central area of research. Efficient management and processing of this massively exponential amount of data from diverse sources, such as telecommunication call data records, telescope imagery, online transaction records, web pages, stock markets, medical records (monitoring critical health conditions of patients), climate warning systems, etc., has become a necessity. Removing redundancy from such huge (multi-billion records) datasets results in resource and compute efficiency for downstream processing and constitutes an important area of study. "Intelligent compression" or deduplication in streaming scenarios, for precise identification and elimination of duplicates from the unbounded data stream is a greater challenge given the real-time nature of data arrival. Stable Bloom Filters (SBF) [13] address this problem to a certain extent. However, SBF suffers from a high false negative rate and slow convergence rate, thereby rendering it inefficient for applications with low false negative rate tolerance.

Petascale Analytics is a hot research area both in academia and industry. It envisages processing massive amounts of data at extremely high rates to generate new scientific insights along with positive impact (for both users and providers) of industries such as E-commerce, Telecom, Finance, Life Sciences and so forth. We consider collaborative filtering (CF) and Clustering algorithms that are key fundamental analytics kernels that help in achieving these aims. Real-time CF and co-clustering on highly sparse massive datasets, while achieving a high prediction accuracy, is a computationally challenging problem. In this paper, we present a novel hierarchical design for soft real-time (less than 1 minute.) distributed co-clustering based collaborative filtering algorithm. Our distributed algorithm has been optimized for multi-core cluster architectures. Theoretical analysis of the time complexity of our algorithm proves the efficacy of our approach. Using the Netflix dataset (900M training ratings with replication) as well as the Yahoo KDD Cup 1 (4.6B training ratings with replication) datasets, we demonstrate the performance and scalability of our algorithm on a 4096-node multi-core cluster architecture. Our distributed algorithm (implemented using OpenMP with MPI) demonstrates around 4x better performance (on Blue Gene/P) as compared to the best prior work, along with high accuracy (26 ± 4 RMSE for Yahoo KDD Cup data and 0.87 ± 0.02 for Netflix data). To the best of our knowledge, these are the best known performance results for collaborative filtering, at high prediction accuracy, for multi-core cluster architectures.

Next Generation Sequencing (NGS) platforms typically produce short reads of size 50-150 base pairs (bp). The number of such short reads can be up to 6 billion per run. To align these short reads to a large genome is a computationally challenging problem. In this paper, we address this problem by considering the design and optimization of parallel sequence alignment on GPU based hybrid architectures. Even though the sequence alignment algorithm is inherently data-parallel, issues such as (a) space-time trade-offs in the Indexing schema, (b) need for fast candidate location search (CAL) on GPU, (c) maintaining low divergence along with low space for the dynamic programming based local alignment, make this a very challenging problem. We present the design of our novel parallel algorithm Graphics processor Accelerated BFAST (GrABFAST) for large scale read alignment that overcomes these challenges and demonstrates superior performance compared to Intel multi-core architectures. Using 5 large genomes including those of Humans, Maize, Horse, Dog and Bacteria, we demonstrate a speedup of around 6x using Fermi Tesla C2070 GPUs vs the BFAST algorithm on 16 core Intel Xeon 5570 architecture.

Big data analytics is a hot research area both in academia and industry. It envisages processing massive amounts of data at high rates to generate new insights leading to positive impact (for both users and providers) of industries such as E-commerce, Telecom, Finance, Life Sciences and so forth. We consider collaborative filtering (CF) and Clustering algorithms that are key fundamental analytics kernels that help in achieving these aims. High throughput CF and co-clustering on highly sparse and massive datasets, along with a high prediction accuracy, is a computationally challenging problem. In this paper, we present a novel hierarchical design for soft real-time (less than 1-minute.) distributed co-clustering based collaborative filtering algorithm. We study both the online and offline variants of this algorithm. Theoretical analysis of the time complexity of our algorithm proves the efficacy of our approach. Further, we present the impact of load balancing based optimizations on multi-core cluster architectures. Using the Netflix dataset(900M training ratings with replication) as well as the Yahoo KDD Cup(2.3B training ratings with replication) datasets, we demonstrate the performance and scalability of our algorithm on a large multi-core cluster architecture. In offline mode, our distributed algorithm demonstrates around 4x better performance (on Blue Gene/P) as compared to the best prior work, along with high accuracy. In online mode, we demonstrated around 3x better performance compared to baseline MPI implementation. To the best of our knowledge, our algorithm provides the best known online and offline performance and scalability results with high accuracy on multi-core cluster architectures.

Removing redundancy in the data is an important problem as it helps in resource and compute efficiency for downstream processing of massive (10 million to 100 million records) datasets. In application domains such as IR, stock markets, telecom and others there is a strong need for real-time data redundancy removal of enormous amounts of data flowing at the rate of 1Gb/s or higher. We consider the problem of finding Range Motifs (clusters) over records in a large dataset such that records within the same cluster are approximately close to each other. This problem is closely related to the approximate nearest neighbour search but is more computationally expensive. Real-time scalable approximate Range Motif discovery on massive datasets is a challenging problem. We present the design of novel sequential and parallel approximate Range Motif discovery and data de-duplication algorithms using Bloom filters. We establish asymptotic upper bounds on the false positive and false negative rates for our algorithm. Further, time complexity analysis of our parallel algorithm on multi-core architectures has been presented. For 10 million records, our parallel algorithm can perform approximate Range Motif discovery and data de-duplication, on 4 sets (clusters), in 59s, on 16 core Intel Xeon 5570 architecture. This gives a throughput of around 170K records/s and around 700Mb/s (using records of size 4K bits). To the best of our knowledge, this is the highest real-time throughput for approximate Range Motif discovery and data redundancy removal on such massive datasets.

The ever growing need to process and analyze massive amounts of data from diverse sources such as telecom call data records, telescope imagery, web pages, stock markets, medical records and other domains has triggered worldwide research in data intensive computing. A key requirement here involves removing redundancy from data, as this enhances the compute efficiency for downstream data processing. These application domains have an intense need for high throughput data deduplication for huge volumes of data flowing at the rate of 1 GB/s or more. In this paper, we present the design of a novel parallel data redundancy removal algorithm. We also present a queueing theoretic analysis to optimize the throughput of our parallel algorithm on multi-core architectures. For 500M records, our parallel algorithm can perform complete deduplication in 255s, on 16 core Intel Xeon 5570 architecture. This gives a throughput of around 2M records/s. For 2048 byte records, we achieve a throughput of 0.81 GB/s. To the best of our knowledge, this is the highest throughput for data redundancy removal on such massive datasets. We also demonstrate strong and weak scalability of our algorithm for both multi-core Power6 and Intel Xeon 5570 architectures.

In this paper, we present the design of a novel scalable parallel algorithm for community detection optimized for multi-core and GPU architectures. Our algorithm is based on label propagation, which works solely on local information, thus giving it the scalability advantage over conventional approaches. We also show that weighted label propagation can overcome typical quality issues in communities detected with label propagation. Experimental results on well known massive scale graphs such as Wikipedia (100M edges) and also on RMAT graphs with 10M 40M edges, demonstrate the superior performance and scalability of our algorithm compared to the well known approaches for community detection. On the hep-th graph (352K edges) and the wikipedia graph (100M edges), using Power 6 architecture with 32 cores, our algorithm achieves one to two orders of magnitude better performance compared to the best known prior results on parallel architectures with similar number of CPUs. Further, our GPGPU based algorithm achieves 8× improvement over the Power 6 performance on 40M edge R-MAT graph. Alongside, we achieve high quality (modularity) of communities detected, with experimental evidence from well-known graphs such as Zachary karate club, Dolphin network and Football club, where we achieve modularity that is close to the best known alternatives. To the best of our knowledge these are best known results for community detection on massive graphs (100M edges) in terms of performance and also quality vs. performance trade-off. This is also a unique work on community detection on GPGPUs with scalable performance.

Data intensive computing has become a central theme in research community and industry. There is an ever growing need to process and analyze massive amounts of data from diverse sources such as telecom call data records, telescope imagery, online transaction records, web pages, stock markets, medical records (monitoring critical health conditions of patients), climate warning systems, etc. Removing redundancy in the data is an important problem as it helps in resource and compute efficiency for downstream processing of the massive (1 billion to 10 billion records) datasets. In application domains such as IR, stock markets, telecom and others, there is a strong need for real-time data redundancy removal (referred to as DRR) of enormous amounts of data flowing at the rate of 1 GB/s or more. Real-time scalable data redundancy removal on massive datasets is a challenging problem. We present the design of a novel parallel data redundancy removal algorithm for both in-memory and disk-based execution. We also develop queueing theoretic analysis to optimize the throughput of our parallel algorithm on multi-core architectures. For 500 million records, our parallel algorithm can perform complete de-duplication in 255s, on 16 core Intel Xeon 5570 architecture, with in-memory execution. This gives a throughput of 2M records/s. For 6 billion records, our parallel algorithm can perform complete de-duplication in less than 4.5 hours, using 6 cores of Intel Xeon 5570, with disk-based execution. This gives a throughput of around 370K records/s. To the best of our knowledge, this is the highest real-time throughput for data redundancy removal on such massive datasets. We also demonstrate the scalability of our algorithm with increasing number of cores and data.

Collaborative filtering (CF) based recommender systems have gained wide popularity in Internet companies like Amazon, Netflix, Google News, and others. These systems make automatic predictions about the interests of a user by inferring from information about like-minded users. Realtime CF on highly sparse massive datasets, while achieving a high prediction accuracy, is a computationally challenging problem. In this paper, we present the design of a soft real-time (around 1 min.) parallel CF algorithm based on the Concept Decomposition technique. Our parallel algorithm has been optimized for multicore/many-core architectures while maintaining the prediction accuracy of 0.84 RMSE. Using the Netflix dataset, we demonstrate the performance and scalability of our algorithm (in both batch mode and online mode) on a 32-core Power6 based SMP system. Our parallel algorithm delivered training time of 64s on the full Netflix dataset and prediction time of 4.5s on 1.4M ratings (3.2/μs per rating prediction). This is 12.6× better than the best known sequential training time and around 33 × better than the best known sequential prediction time, along with high accuracy (0.84 RMSE). To the best of our knowledge, this is also the best known parallel performance at such high accuracy.

Stream computing research is moving from terascale to petascale levels. It aims to rapidly analyze data as it streams in from many sources and make decisions with high speed and accuracy in fields as diverse as security surveillance and financial services including stock trading. We specifically consider real-time text indexing and search with high input data rates (10 GB/s or more) along with small index age-off (expiry) time. This makes it necessary to have maximal indexing rates for large volumes of data as well as minimal latency for indexing (time between start of indexing for a document and its availability for search) while maintaining very-low search response time. In addition, future massively parallel architectures with storage class memories will enable high speed in-memory real-time indexing, where index can be completely stored in a high capacity storage class memory. In this paper, we present the design of distributed data-structures and distributed real-time text indexing algorithm for parallel systems having large (thousands to hundred thousand) number of cores/processors, while simultaneously providing acceptable search performance [1]. The inherent trade-offs involved in index space, indexing throughput and search response time make this problem particularly challenging. Our algorithm uses group-based index construction and leverages novel index data structures that reduce load imbalance and make text indexing and merge process more scalable and efficient. We show analytically that the asymptotic parallel time complexity of our distributed indexing algorithm, is at least ?(log(P)) factor better than typical indexing approaches, where P is the number of indexing nodes in a group. We further demonstrate the performance and scalability of our distributed indexing algorithm, on an MPP architecture (Blue Gene/L) using actual IBM intranet data. We achieved high indexing throughput of around 312 GB/min on an 8 K node Blue Gene/L machine. In comparison with parallel indexing implemented using typical approaches like CLucene, this is 3?-7? better. To the best of our knowledge, this is the first published result on indexing throughput at such a large scale, with sustained search performance. We further show that our approach is scalable to 128 K nodes, giving an estimated indexing throughput of 5 T B/min. We also achieved indexing latency that is around 10? better than typical indexing approaches.

DISC (Data-Intensive Super Computing) is gaining strong research momentum [1]. DISC systems differ from conventional supercomputers in their focus on data, as they acquire and maintain continually changing massive data sets, in addition to performing large-scale computations over the data. Towards this end, we consider the problem of real-time text indexing and search with high input data rates (10 GB/s or more) along with small index age-off time while sustaining search response time. Load imbalance and communication bottlenecks make this problem particularly challenging. We present performance optimizations for distributed in-memory text indexing of massive input data sets on parallel systems having large number of cores/processors, with sustained search performance [2]. Our distributed indexing algorithm uses a hybrid group-based approach which enables scalable indexing and search over massively parallel systems. In addition, we designed and analyzed communication optimizations including routing using Steiner nodes and topology mapping. Using theoretical analysis for the asymptotic parallel time complexity we establish the scalability of our algorithm with |P| for small index-group size and scalability with |C| for larger index-group size; where |P| and |C| are the number of Producers and Consumers respectively, in the index-group. We have obtained indexing throughput of 524 GB/min on 4K nodes of Blue Gene/L1using actual IBM intranet data, which is 3.36x better than the previous best throughput [3] and 10.3x better than typical indexing approaches such as CLucene2on the same number of nodes. This gives an estimated throughput of 17 TB/min on 128K nodes with sustained search performance. We also demonstrate the improvements in strong and weak scalability of our distributed indexing algorithm over the previous best [3] on Blue Gene/L. To the best of our knowledge, this is the highest indexing throughput ever published on a large scale system with sustained search performance.