E2E Model Research Articles

Multi-span optical power spectrum prediction using cascaded learning with one-shot end-to-end measurement

Scalable methods for optical transmission performance prediction using machine learning (ML) are studied in metro reconfigurable optical add-drop multiplexer (ROADM) networks. A cascaded learning framework is introduced to encompass the use of cascaded component models for end-to-end (E2E) optical path prediction augmented with different combinations of E2E performance data and models. Additional E2E optical path data and models are used to reduce the prediction error accumulation in the cascade. Off-line training (pre-trained prior to deployment) and transfer learning are used for component-level erbium-doped fiber amplifier (EDFA) gain models to ensure scalability. Considering channel power prediction, we show that the data collection process of the pre-trained EDFA model can be reduced to only 5% of the original training set using transfer learning. We evaluate the proposed method under three different topologies with field deployed fibers and achieve a mean absolute error of 0.16 dB with a single (one-shot) E2E measurement on the deployed 6-span system with 12 EDFAs.

HubNet: An E2E Model for Wheel Hub Text Detection and Recognition Using Global and Local Features.

Automatic detection and recognition of wheel hub text, which can boost the efficiency and accuracy of product information recording, are undermined by the obscurity and orientation variability of text on wheel hubs. To address these issues, this paper constructs a wheel hub text dataset and proposes a wheel hub text detection and recognition model called HubNet. The dataset captured images on real industrial production line scenes, including 446 images, 934 word instances, and 2947 character instances. HubNet is an end-to-end text detection and recognition model, not only comprising conventional detection and recognition heads but also incorporating a feature cross-fusion module, which improves the accuracy of recognizing wheel hub texts by utilizing both global and local features. Experimental results show that on the wheel hub text dataset, the HubNet achieves an accuracy of 86.5%, a recall of 79.4%, and an F1-score of 0.828, and the feature cross-fusion module increases the accuracy by 2% to 4%. The wheel hub dataset and the HubNet offer a significant reference for automatic detection and recognition of wheel hub text.

Decoupled structure for improved adaptability of end-to-end models

Although end-to-end (E2E) trainable automatic speech recognition (ASR) has shown great success by jointly learning acoustic and linguistic information, it still suffers from the effect of domain shifts, thus limiting potential applications. The E2E ASR model implicitly learns an internal language model (LM) which characterises the training distribution of the source domain, and the E2E trainable nature makes the internal LM difficult to adapt to the target domain with text-only data. To solve this problem, this paper proposes decoupled structures for attention-based encoder–decoder (Decoupled-AED) and neural transducer (Decoupled-Transducer) models, which can achieve flexible domain adaptation in both offline and online scenarios while maintaining robust intra-domain performance. To this end, the acoustic and linguistic parts of the E2E model decoder (or prediction network) are decoupled, making the linguistic component (i.e. internal LM) replaceable. When encountering a domain shift, the internal LM can be directly replaced during inference by a target-domain LM, without re-training or using domain-specific paired speech-text data. Experiments for E2E ASR models trained on the LibriSpeech-100h corpus showed that the proposed decoupled structure gave 15.1% and 17.2% relative word error rate reductions on the TED-LIUM 2 and AESRC2020 corpora while still maintaining performance on intra-domain data. It is also shown that the decoupled structure can be used to boost cross-domain speech translation quality while retaining the intra-domain performance.

Designing a Novel Learning Approach for Infectious Lung Disease Prediction using Bi-Directional Recurrent Neural Network (BRNN)

The principal goal of this research is to use regular imaging to diagnose and classify several lung disorders, including pneumonia, COVID-19 and tuberculosis.A novel Bi-directional Recurrent Neural Network (BRNN) model is implemented in this work to categorize the disease according to the input samples given. With the help of the feature extractor, prediction model, and classification model, the suggested model performs feature analysis. The E2E model that has been suggested carries out knowledge acquisition using an efficient computational and memory approach. The benchmark dataset, made accessible online, was used to test the proposed model. When networks are trained to extract characteristics, the developments in deep learning techniques yield encouraging results and offer higher potential efficiency than biomedical applications. The main goal is to evaluate the significance of current methods and validate the BRNN model to address the concerns that are currently present. The suggested model offers better efficiency and performance for better detection. The results reveal that the model with a recurrent network works better than the current methods, with an accuracy of 71% for all diseases, whereas the traditional network models perform worse. The suggested model has less computational overhead and fewer training parameters. When compared to alternative methods, the model offers a superior trade-off.

End-to-End Top-Down Load Forecasting Model for Residential Consumers

This study presents an efficient end-to-end (E2E) learning approach for the short-term load forecasting of hierarchically structured residential consumers based on the principles of a top-down (TD) approach. This technique employs a neural network for predicting load at lower hierarchical levels based on the aggregated one at the top. A simulation is carried out with 9 (from 2013 to 2021) years of energy consumption data of 50 houses located in the United States of America. Simulation results demonstrate that the E2E model, which uses a single model for different nodes and is based on the principles of a top-down approach, shows huge potential for improving forecasting accuracy, making it a valuable tool for grid planners. Model inputs are derived from the aggregated residential category and the specific cluster targeted for forecasting. The proposed model can accurately forecast any residential consumption cluster without requiring any hyperparameter adjustments. According to the experimental analysis, the E2E model outperformed a two-stage methodology and a benchmarked Seasonal Autoregressive Integrated Moving Average (SARIMA) and Support Vector Regression (SVR) model by a mean absolute percentage error (MAPE) of 2.27%.

Joint speaker encoder and neural back-end model for fully end-to-end automatic speaker verification with multiple enrollment utterances

Conventional automatic speaker verification systems can usually be decomposed into a front-end model such as time delay neural network (TDNN) for extracting speaker embeddings and a back-end model such as statistics-based probabilistic linear discriminant analysis (PLDA) or neural network-based neural PLDA (NPLDA) for similarity scoring. However, the sequential optimization of the front-end and back-end models may lead to a local minimum, which theoretically prevents the whole system from achieving the best optimization. Although some methods have been proposed for jointly optimizing the two models, such as the generalized end-to-end (GE2E) model and NPLDA E2E model, most of these methods have not fully investigated how to model the intra-relationship between multiple enrollment utterances. In this paper, we propose a new E2E joint method for speaker verification especially designed for the practical scenario of multiple enrollment utterances. To leverage the intra-relationship among multiple enrollment utterances, our model comes equipped with frame-level and utterance-level attention mechanisms. Additionally, focal loss is utilized to balance the importance of positive and negative samples within a mini-batch and focus on the difficult samples during the training process. We also utilize several data augmentation techniques, including conventional noise augmentation using MUSAN and RIRs datasets and a unique speaker embedding-level mixup strategy for better optimization.

Abstract B016: AI analysis of histological images accurately identifies luminal subtype urothelial carcinomas characterized by high PPARG expression

Abstract Background PPARG is a cell lineage determining transcription factor in muscle invasive urothelial carcinoma (MIUC), where high expression is associated with the luminal subtype (1, 2). As FX-909(2), a PPARG inverse agonist, enters the clinic, biomarkers that reflect the luminal subtype will reveal patients with the potential to respond to PPARG inhibition. The determination of luminal status is generally performed via RNA-seq and/or multiple immunohistochemistry stains, which are costly and time-consuming. However, MIUC biopsies are routinely stained with hematoxylin and eosin (H&E). Machine learning (ML)-driven analyses of H&E-stained tissue may enable the identification of patients with luminal MIUC and have advantages over the current molecular approach. Methods H&E-stained slides from 367 unique primary MIUC cases from the TCGA BLCA dataset were split into training (70%), validation (15%), and held-out test (15%) sets by preserving the data distribution of patient metadata. A curated retrospective cohort of 42 localized, stage III-IV primary MIUCs was used as an independent test set. Molecular classification as luminal (luminal papillary, luminal, and luminal infiltrated subtypes) or non-luminal (basal-squamous and neuronal subtypes) was performed and used as ground truth(3). Pretrained artifact and tissue segmentation models were deployed on all images to identify artifact-free areas of cancer and cancer-associated stroma. An end-to-end (E2E) additive multiple instance learning model was trained to identify luminal cases using the training set. Top performing model iterations were compared on the validation set, and the optimal iteration was deployed on both test sets. Results We assessed the performance of our E2E model in predicting luminal status using the molecular subtypes derived from Robertson et al. as ground truth(3). The E2E model showed excellent performance when predicting luminal status in the TCGA validation, TCGA test, and independent test sets (AUROC = 0.96, 0.95, and 0.97, respectively). The accuracy in all three cohorts was 89-90%, with a sensitivity of 0.86-0.96, a specificity of 0.82-0.94, and an F1 score of 0.88-0.9. Conclusions We generated a robust ML model that accurately predicts luminal MIUC using H&E-stained slides. Luminal MIUC is dependent on PPARG, and PPARG inverse agonism represents a promising therapeutic approach for MIUC. Coupled with the first-in-class FX-909 therapeutic entering the clinic, the strong performance of our model highlights the potential for its application as a precision biomarker to identify patients with advanced urothelial carcinoma likely to respond to PPARG inhibition.

An end-to-end deep learning model for solving data-driven newsvendor problem with accessibility to textual review data

We investigate a data-driven single-period inventory management problem with uncertain demand, where large amounts of textual online reviews and historical data are accessible. Unlike two-step frameworks (i.e., predict-then-optimization), we propose an end-to-end (E2E) framework that directly suggests the order quantity by leveraging a deep learning model that inputs textual online reviews and other demand-related feature data, without any intermediate steps such as text sentiment analysis. The E2E model does not require any prior assumptions about the demand distribution and can automatically determine the order quantity that minimizes the newsvendor cost by employing the information from real-world data. Our experiments, using publicly available real-world data, demonstrate that our method can significantly reduce the sum of overage and underage costs, outperforming other data-driven models proposed in recent years. Specifically, the inclusion of textual online review data improves ordering decisions by a 28.7% cost reduction.

ColBERT-PRF: Semantic Pseudo-Relevance Feedback for Dense Passage and Document Retrieval

Pseudo-relevance feedback mechanisms, from Rocchio to the relevance models, have shown the usefulness of expanding and reweighting the users’ initial queries using information occurring in an initial set of retrieved documents, known as the pseudo-relevant set. Recently, dense retrieval – through the use of neural contextual language models such as BERT for analysing the documents’ and queries’ contents and computing their relevance scores – has shown a promising performance on several information retrieval tasks still relying on the traditional inverted index for identifying documents relevant to a query. Two different dense retrieval families have emerged: the use of single embedded representations for each passage and query, e.g., using BERT’s [CLS] token, or via multiple representations, e.g., using an embedding for each token of the query and document (exemplified by ColBERT). In this work, we conduct the first study into the potential for multiple representation dense retrieval to be enhanced using pseudo-relevance feedback and present our proposed approach ColBERT-PRF. In particular, based on the pseudo-relevant set of documents identified using a first-pass dense retrieval, ColBERT-PRF extracts the representative feedback embeddings from the document embeddings of the pseudo-relevant set. Among the representative feedback embeddings, the embeddings that most highly discriminate among documents are employed as the expansion embeddings, which are then added to the original query representation. We show that these additional expansion embeddings both enhance the effectiveness of a reranking of the initial query results as well as an additional dense retrieval operation. Indeed, experiments on the MSMARCO passage ranking dataset show that MAP can be improved by up to 26% on the TREC 2019 query set and 10% on the TREC 2020 query set by the application of our proposed ColBERT-PRF method on a ColBERT dense retrieval approach.We further validate the effectiveness of our proposed pseudo-relevance feedback technique for a dense retrieval model on MSMARCO document ranking and TREC Robust04 document ranking tasks. For instance, ColBERT-PRF exhibits up to 21% and 14% improvement in MAP over the ColBERT E2E model on the MSMARCO document ranking TREC 2019 and TREC 2020 query sets, respectively. Additionally, we study the effectiveness of variants of the ColBERT-PRF model with different weighting methods. Finally, we show that ColBERT-PRF can be made more efficient, attaining up to 4.54× speedup over the default ColBERT-PRF model, and with little impact on effectiveness, through the application of approximate scoring and different clustering methods.

End-to-End Model-Based Detection of Infants with Autism Spectrum Disorder Using a Pretrained Model

In this paper, we propose an end-to-end (E2E) neural network model to detect autism spectrum disorder (ASD) from children's voices without explicitly extracting the deterministic features. In order to obtain the decisions for discriminating between the voices of children with ASD and those with typical development (TD), we combined two different feature-extraction models and a bidirectional long short-term memory (BLSTM)-based classifier to obtain the ASD/TD classification in the form of probability. We realized one of the feature extractors as the bottleneck feature from an autoencoder using the extended version of the Geneva minimalistic acoustic parameter set (eGeMAPS) input. The other feature extractor is the context vector from a pretrained wav2vec2.0-based model directly applied to the waveform input. In addition, we optimized the E2E models in two different ways: (1) fine-tuning and (2) joint optimization. To evaluate the performance of the proposed E2E models, we prepared two datasets from video recordings of ASD diagnoses collected between 2016 and 2018 at Seoul National University Bundang Hospital (SNUBH), and between 2019 and 2021 at a Living Lab. According to the experimental results, the proposed wav2vec2.0-based E2E model with joint optimization achieved significant improvements in the accuracy and unweighted average recall, from 64.74% to 71.66% and from 65.04% to 70.81%, respectively, compared with a conventional model using autoencoder-based BLSTM and the deterministic features of the eGeMAPS.

A Practical End-to-End Inventory Management Model with Deep Learning

We investigate a data-driven multiperiod inventory replenishment problem with uncertain demand and vendor lead time (VLT) with accessibility to a large quantity of historical data. Different from the traditional two-step predict-then-optimize (PTO) solution framework, we propose a one-step end-to-end (E2E) framework that uses deep learning models to output the suggested replenishment amount directly from input features without any intermediate step. The E2E model is trained to capture the behavior of the optimal dynamic programming solution under historical observations without any prior assumptions on the distributions of the demand and the VLT. By conducting a series of thorough numerical experiments using real data from one of the leading e-commerce companies, we demonstrate the advantages of the proposed E2E model over conventional PTO frameworks. We also conduct a field experiment with JD.com, and the results show that our new algorithm reduces holding cost, stockout cost, total inventory cost, and turnover rate substantially compared with JD’s current practice. For the supply chain management industry, our E2E model shortens the decision process and provides an automatic inventory management solution with the possibility to generalize and scale. The concept of E2E, which uses the input information directly for the ultimate goal, can also be useful in practice for other supply chain management circumstances. This paper was accepted by Hamid Nazerzadeh, big data analytics. Funding: This research was supported by the National Key Research and Development Program of China [Grant 2018YFB1700600] and National Natural Science Foundation of China [Grants 71991462 and 91746210]. Supplemental Material: The online data are available at https://doi.org/10.1287/mnsc.2022.4564 .

End-to-End Deep Convolutional Recurrent Models for Noise Robust Waveform Speech Enhancement.

Because of their simple design structure, end-to-end deep learning (E2E-DL) models have gained a lot of attention for speech enhancement. A number of DL models have achieved excellent results in eliminating the background noise and enhancing the quality as well as the intelligibility of noisy speech. Designing resource-efficient and compact models during real-time processing is still a key challenge. In order to enhance the accomplishment of E2E models, the sequential and local characteristics of speech signal should be efficiently taken into consideration while modeling. In this paper, we present resource-efficient and compact neural models for end-to-end noise-robust waveform-based speech enhancement. Combining the Convolutional Encode-Decoder (CED) and Recurrent Neural Networks (RNNs) in the Convolutional Recurrent Network (CRN) framework, we have aimed at different speech enhancement systems. Different noise types and speakers are used to train and test the proposed models. With LibriSpeech and the DEMAND dataset, the experiments show that the proposed models lead to improved quality and intelligibility with fewer trainable parameters, notably reduced model complexity, and inference time than existing recurrent and convolutional models. The quality and intelligibility are improved by 31.61% and 17.18% over the noisy speech. We further performed cross corpus analysis to demonstrate the generalization of the proposed E2E SE models across different speech datasets.

Reliability-Based Large-Vocabulary Audio-Visual Speech Recognition.

Audio-visual speech recognition (AVSR) can significantly improve performance over audio-only recognition for small or medium vocabularies. However, current AVSR, whether hybrid or end-to-end (E2E), still does not appear to make optimal use of this secondary information stream as the performance is still clearly diminished in noisy conditions for large-vocabulary systems. We, therefore, propose a new fusion architecture—the decision fusion net (DFN). A broad range of time-variant reliability measures are used as an auxiliary input to improve performance. The DFN is used in both hybrid and E2E models. Our experiments on two large-vocabulary datasets, the Lip Reading Sentences 2 and 3 (LRS2 and LRS3) corpora, show highly significant improvements in performance over previous AVSR systems for large-vocabulary datasets. The hybrid model with the proposed DFN integration component even outperforms oracle dynamic stream-weighting, which is considered to be the theoretical upper bound for conventional dynamic stream-weighting approaches. Compared to the hybrid audio-only model, the proposed DFN achieves a relative word-error-rate reduction of 51% on average, while the E2E-DFN model, with its more competitive audio-only baseline system, achieves a relative word error rate reduction of 43%, both showing the efficacy of our proposed fusion architecture.

Code-switched end-to-end Marathi speech recognition for especially abled people

There is no need for eyes or arms in speech communication. Instead, both humans and machines communicate by speaking and listening. Therefore, speech communication not only helps physically disabled but also help people with hearing loss and/or language disorders. It is also used by people with vision or motor disabilities. There are various types of devices that are used for people with hearing loss. They can be classified as audio-visual stimulation devices, tactile devices, and speech processing devices. These devices consider the characteristics of speech and/or deploy automatic speech recognition inside it. This article focuses on such ASR system based on E2E model which can be useful in such devices. This research is carried out on the language used by the families of India’s over 80 million speakers i.e., Marathi. These assessments provide valuable insights into the development of SR models for low-resourced Indian languages. We present E2E ASR model trained with code-switched text of Marathi language and decoded similarly to a normal E2E ASR system. This model achieves a better word error rate of 35.09%, according to experimental results on the code-switched Marathi corpus. We compared our model to Facebook’s wav2vec2 model, which produces a WER of 23.72 percent on code-switched Marathi datasets without any language model and 12.10 percent on open ASR Marathi datasets.

End-to-End Spoken Language Understanding: Performance analyses of a voice command task in a low resource setting

Spoken Language Understanding (SLU) is a core task in most human–machine interaction systems . With the emergence of smart homes, smart phones and smart speakers, SLU has become a key technology for the industry. In a classical SLU approach, an Automatic Speech Recognition (ASR) module transcribes the speech signal into a textual representation from which a Natural Language Understanding (NLU) module extracts semantic information. Recently End-to-End SLU (E2E SLU) based on Deep Neural Networks has gained momentum since it benefits from the joint optimisation of the ASR and the NLU parts, hence limiting the cascade of error effect of the pipeline architecture. However, little is known about the actual linguistic properties used by E2E models to predict concepts and intents from speech input. In this paper, we present a study identifying the signal features and other linguistic properties used by an E2E model to perform the SLU task. The study is carried out in the application domain of a smart home that has to handle non-English (here French) voice commands.The results show that a good E2E SLU performance does not always require a perfect ASR capability. Furthermore, the results show the superior capabilities of the E2E model in handling background noise and syntactic variation compared to the pipeline model. Finally, a finer-grained analysis suggests that the E2E model uses the pitch information of the input signal to identify voice command concepts. The results and methodology outlined in this paper provide a springboard for further analyses of E2E models in speech processing.

Improving low-resource Tibetan end-to-end ASR by multilingual and multilevel unit modeling

Conventional automatic speech recognition (ASR) and emerging end-to-end (E2E) speech recognition have achieved promising results after being provided with sufficient resources. However, for low-resource language, the current ASR is still challenging. The Lhasa dialect is the most widespread Tibetan dialect and has a wealth of speakers and transcriptions. Hence, it is meaningful to apply the ASR technique to the Lhasa dialect for historical heritage protection and cultural exchange. Previous work on Tibetan speech recognition focused on selecting phone-level acoustic modeling units and incorporating tonal information but underestimated the influence of limited data. The purpose of this paper is to improve the speech recognition performance of the low-resource Lhasa dialect by adopting multilingual speech recognition technology on the E2E structure based on the transfer learning framework. Using transfer learning, we first establish a monolingual E2E ASR system for the Lhasa dialect with different source languages to initialize the ASR model to compare the positive effects of source languages on the Tibetan ASR model. We further propose a multilingual E2E ASR system by utilizing initialization strategies with different source languages and multilevel units, which is proposed for the first time. Our experiments show that the performance of the proposed method-based ASR system exceeds that of the E2E baseline ASR system. Our proposed method effectively models the low-resource Lhasa dialect and achieves a relative 14.2% performance improvement in character error rate (CER) compared to DNN-HMM systems. Moreover, from the best monolingual E2E model to the best multilingual E2E model of the Lhasa dialect, the system’s performance increased by 8.4% in CER.

Self-Supervised Pre-Training for Attention-Based Encoder-Decoder ASR Model

End-to-end (E2E) models, including the attention-based encoder-decoder (AED) models, have achieved promising performance on the automatic speech recognition (ASR) task. However, the supervised training process of the E2E model needs a large amount of speech-text paired data. In contrast, self-supervised pre-training can pre-train the model on the unlabeled data and then fine-tune it on the limited labeled data to realize better performance. Most of the previous self-supervised pre-training methods focus on learning hidden representations from speech but ignore how to utilize the unpaired text. As a result, previous works often pre-train an acoustic encoder and then fine-tune it as a classification based ASR model, such as Connectionist Temporal Classification (CTC) based model, rather than an AED model. In this paper, we propose a self-supervised pre-training method for the AED model (SP-AED). The SP-AED method contains acoustic pre-training for the encoder, linguistic pre-training for the decoder, and an adaptive combination fine-tuning for the whole system. We first design a linguistic pre-training method for decoder by utilizing the text-only data. The decoder will be pre-trained as a noise-condition language model to learn the prior distribution of the text. Then, we pre-train the AED encoder with the wav2vec2.0 method with some modifications. Finally, we combine the pre-trained encoder and decoder and fine-tune them on the limited labeled data. We design an adaptive combination method during fine-tuning by modifying the decoder's input and output to prevent catastrophic forgetting. Experiments prove that compared with the random initialized models, the SP-AED pre-trained models can realize up to 17% relative improvement. And with similar model size or computational cost, we can get comparable results to other classification-based models on both English and Chinese corpus.

An E2E-ASR-Based Iteratively-Trained Timestamp Estimator

Text-to-speech alignment, also known as time alignment, is essential for automatic speech recognition (ASR) systems used for speech retrieval tasks, such as keyword search and speech segment extraction. Previous works have used the Gaussian mixture model-hidden Markov model (GMM-HMM) forced alignment to improve the alignment performance. However, when used with end-to-end (E2E) ASR, GMM-HMM forced alignment causes extra reliance on expertise such as pronunciation lexica. It also increases the system complexity because GMM-HMMs are very dissimilar to E2E models. To tackle these two problems, we propose an E2E-ASR-based iteratively-trained timestamp estimator (ITSE), which performs alignment between token-level transcription and speech. We train ITSE first with coarse initial alignment targets generated using connectionist temporal classification (CTC) posteriors. During training, we iteratively perform realignment to update the targets. We attribute the effectiveness of the iterative training to ITSE’s two vital features. First, ITSE performs alignment using similarities between token and speech embeddings instead of frame-wise token classification posteriors. Second, ITSE uses speech embeddings that are aware of left context rather than global context. ITSE significantly outperforms CTC-based baselines in word alignment accuracy and is comparable to a GMM-HMM forced aligner. In short, ITSE is an accurate, lightweight text-to-speech alignment module implemented without expertise such as pronunciation lexica.

A Hierarchy-Based System for Recognizing Customer Activity in Retail Environments

Customer activity (CA) in retail environments, which ranges over various shopper situations in store spaces, provides valuable information for store management and marketing planning. Several systems have been proposed for customer activity recognition (CAR) from in-store camera videos, and most of them use machine learning based end-to-end (E2E) CAR models, due to their remarkable performance. Usually, such E2E models are trained for target conditions (i.e., particular CA types in specific store spaces). Accordingly, the existing systems are not malleable to fit the changes in target conditions because they require entire retraining of their specialized E2E models and concurrent use of additional E2E models for new target conditions. This paper proposes a novel CAR system based on a hierarchy that organizes CA types into different levels of abstraction from lowest to highest. The proposed system consists of multiple CAR models, each of which performs CAR tasks that belong to a certain level of the hierarchy on the lower level’s output, and thus conducts CAR for videos through the models level by level. Since these models are separated, this system can deal efficiently with the changes in target conditions by modifying some models individually. Experimental results show the effectiveness of the proposed system in adapting to different target conditions.

Exploring Transfer Learning For End-to-End Spoken Language Understanding

Voice Assistants such as Alexa, Siri, and Google Assistant typically use a two-stage Spoken Language Understanding pipeline; first, an Automatic Speech Recognition (ASR) component to process customer speech and generate text transcriptions, followed by a Natural Language Understanding (NLU) component to map transcriptions to an actionable hypothesis. An end-to-end (E2E) system that goes directly from speech to a hypothesis is a more attractive option. These systems were shown to be smaller, faster, and better optimized. However, they require massive amounts of end-to-end training data and in addition, don't take advantage of the already available ASR and NLU training data. In this work, we propose an E2E system that is designed to jointly train on multiple speech-to-text tasks, such as ASR (speech-transcription) and SLU (speech-hypothesis), and text-to-text tasks, such as NLU (text-hypothesis). We call this the Audio-Text All-Task (AT-AT) Model and we show that it beats the performance of E2E models trained on individual tasks, especially ones trained on limited data. We show this result on an internal music dataset and two public datasets, FluentSpeech and SNIPS Audio, where we achieve state-of-the-art results. Since our model can process both speech and text input sequences and learn to predict a target sequence, it also allows us to do zero-shot E2E SLU by training on only text-hypothesis data (without any speech) from a new domain. We evaluate this ability of our model on the Facebook TOP dataset and set a new benchmark for zeroshot E2E performance. We release the audio data collected for the TOP dataset for future research.