Set-valued data are widely used to represent information in the real word, such as individual daily behaviors, items in shopping carts and web browsing history. By collecting set-valued data and identifying heavy hitters, service providers (i.e., the collector) can learn usage preferences of costumers (i.e., users), and improve the quality of their services by the learned information. However, the collection of raw data would bring privacy risks to users. Recently, local differential privacy (LDP) has emerged as a rigorous privacy framework for user private data collection. At the same time, many LDP schemes have been designed to achieve heavy hitters, but most of them are limited by the large data domain due to the huge computation cost. In this paper, we propose an LDP framework: PemSet, to efficiently identify heavy hitters from set-valued data with a large domain. In PemSet, users mainly focus on the prefix of each item (i.e., the first few bits of the binary expression of each item), and only perturb and report prefixes to reduce computation cost. Sometimes the prefixes of different items are the same, so the reported set-valued data could be a multiset, i.e., a set including multiple same items. As such, we design four LDP protocols MOLH, MOLH-S, MPCKV, MWheel to estimate frequencies of items in the multiset setting, and compare their performance under PemSet framework by experiments. Experimental results demonstrate that MOLH can perform the best in a high privacy region, i.e.,$\epsilon < 1$, while MWheel can obtain the highest utility when privacy budget is large, i.e.,$\epsilon \geqslant 1$.

End-to-end spoken language understanding (SLU) recently attracted increasing interest. Compared to the conventional tandem-based approach that combines speech recognition and language understanding as separate modules, the new approach extracts users' intentions directly from the speech signals, resulting in joint optimization and low latency. Such an approach, however, is typically designed to process one intent at a time, which leads users to have to take multiple rounds to fulfill their requirements while interacting with a dialogue system. In this paper, we propose a streaming end-to-end framework that can process multiple intentions in an online and incremental way. The backbone of our framework is a unidirectional RNN trained with the connectionist temporal classification (CTC) criterion. By this design, an intention can be identified when sufficient evidence has been accumulated, and multiple intentions will be identified sequentially. We evaluate our solution on the Fluent Speech Commands (FSC) dataset and the detection accuracy is about 97 % on all multi-intent settings. This result is comparable to the performance of the state-of-the-art non-streaming models, but is achieved in an online and incremental way. We also employ our model to an keyword spotting task using the Google Speech Commands dataset, and the results are also highly promising.

Human-computer interaction (HCI) is significantly impacted by delayed responses from a spoken dialogue system.Hence, endto-end (e2e) spoken language understanding (SLU) solutions have recently been proposed to decrease latency.Such approaches allow for the extraction of semantic information directly from the speech signal, thus bypassing the need for a transcript from an automatic speech recognition (ASR) system.In this paper, we propose a compact e2e SLU architecture for streaming scenarios, where chunks of the speech signal are processed continuously to predict intent and slot values.Our model is based on a 3D convolutional neural network (3D-CNN) and a unidirectional long short-term memory (LSTM).We compare the performance of two alignment-free losses: the connectionist temporal classification (CTC) method and its adapted version, namely connectionist temporal localization (CTL).The latter performs not only the classification but also localization of sequential audio events.The proposed solution is evaluated on the Fluent Speech Command dataset and results show our model ability to process incoming speech signal, reaching accuracy as high as 98.97 % for CTC and 98.78 % for CTL on single-label classification, and as high as 95.69 % for CTC and 95.28 % for CTL on two-label prediction.