Traditionally, research in automated speech recognition has focused on local-first encoding of audio representations to predict the spoken phonemes in an utterance. Unfortunately, approaches relying on such hyper-local information tend to be vulnerable to both local-level corruption (such as audio-frame drops, or loud noises) and global-level noise (such as environmental noise, or background noise) that has not been seen during training. In this work, we introduce a novel approach which leverages a self-supervised learning technique based on masked language modeling to compute a global, multi-modal encoding of the environment in which the utterance occurs. We then use a new deep-fusion framework to integrate this global context into a traditional ASR method, and demonstrate that the resulting method can outperform baseline methods by up to 7% on Librispeech; gains on internal datasets range from 6% (on larger models) to 45% (on smaller models).
Over the last few decades, many aspects of human life have been enhanced with virtual domains, from the advent of digital assistants such as Amazon's Alexa and Apple's Siri to the latest metaverse efforts of the rebranded Meta. These trends underscore the importance of generating photorealistic visual depictions of humans. This has led to the rapid growth of so-called deepfake and talking head generation methods in recent years. Despite their impressive results and popularity, they usually lack certain qualitative aspects such as texture quality, lips synchronization, or resolution, and practical aspects such as the ability to run in real-time. To allow for virtual human avatars to be used in practical scenarios, we propose an end-to-end framework for synthesizing high-quality virtual human faces capable of speech with a special emphasis on performance. We introduce a novel network utilizing visemes as an intermediate audio representation and a novel data augmentation strategy employing a hierarchical image synthesis approach that allows disentanglement of the different modalities used to control the global head motion. Our method runs in real-time, and is able to deliver superior results compared to the current state-of-the-art.
This paper describes the THUEE team's speech recognition system for the IARPA Open Automatic Speech Recognition Challenge (OpenASR21), with further experiment explorations. We achieve outstanding results under both the Constrained and Constrained-plus training conditions. For the Constrained training condition, we construct our basic ASR system based on the standard hybrid architecture. To alleviate the Out-Of-Vocabulary (OOV) problem, we extend the pronunciation lexicon using Grapheme-to-Phoneme (G2P) techniques for both OOV and potential new words. Standard acoustic model structures such as CNN-TDNN-F and CNN-TDNN-F-A are adopted. In addition, multiple data augmentation techniques are applied. For the Constrained-plus training condition, we use the self-supervised learning framework wav2vec2.0. We experiment with various fine-tuning techniques with the Connectionist Temporal Classification (CTC) criterion on top of the publicly available pre-trained model XLSR-53. We find that the frontend feature extractor plays an important role when applying the wav2vec2.0 pre-trained model to the encoder-decoder based CTC/Attention ASR architecture. Extra improvements can be achieved by using the CTC model finetuned in the target language as the frontend feature extractor.
Hate speech is a challenging issue plaguing the online social media. While better models for hate speech detection are continuously being developed, there is little research on the bias and interpretability aspects of hate speech. In this paper, we introduce HateXplain, the first benchmark hate speech dataset covering multiple aspects of the issue. Each post in our dataset is annotated from three different perspectives: the basic, commonly used 3-class classification (i.e., hate, offensive or normal), the target community (i.e., the community that has been the victim of hate speech/offensive speech in the post), and the rationales, i.e., the portions of the post on which their labelling decision (as hate, offensive or normal) is based. We utilize existing state-of-the-art models and observe that even models that perform very well in classification do not score high on explainability metrics like model plausibility and faithfulness. We also observe that models, which utilize the human rationales for training, perform better in reducing unintended bias towards target communities. We have made our code and dataset public at https://github.com/punyajoy/HateXplain
While corpora of child speech and child-directed speech (CDS) have enabled major contributions to the study of child language acquisition, semantic annotation for such corpora is still scarce and lacks a uniform standard. We compile two CDS corpora with sentential logical forms, one in English and the other in Hebrew. In compiling the corpora we employ a methodology that enforces a cross-linguistically consistent representation, building on recent advances in dependency representation and semantic parsing. The corpora are based on a sizable portion of Brown's Adam corpus from CHILDES (about 80% of its child-directed utterances), and to all child-directed utterances from Berman's Hebrew CHILDES corpus Hagar. We begin by annotating the corpora with the Universal Dependencies (UD) scheme for syntactic annotation, motivated by its applicability to a wide variety of domains and languages. We then proceed by applying an automatic method for transducing sentential logical forms (LFs) from UD structures. The two representations have complementary strengths: UD structures are language-neutral and support direct annotation, whereas LFs are neutral as to the interface between syntax and semantics, and transparently encode semantic distinctions. We verify the quality of the annotated UD annotation using an inter-annotator agreement study. We then demonstrate the utility of the compiled corpora through a longitudinal corpus study of the prevalence of different syntactic and semantic phenomena.
Latent Dirichlet allocation (LDA) is widely used for unsupervised topic modelling on sets of documents. No temporal information is used in the model. However, there is often a relationship between the corresponding topics of consecutive tokens. In this paper, we present an extension to LDA that uses a Markov chain to model temporal information. We use this new model for acoustic unit discovery from speech. As input tokens, the model takes a discretised encoding of speech from a vector quantised (VQ) neural network with 512 codes. The goal is then to map these 512 VQ codes to 50 phone-like units (topics) in order to more closely resemble true phones. In contrast to the base LDA, which only considers how VQ codes co-occur within utterances (documents), the Markov chain LDA additionally captures how consecutive codes follow one another. This extension leads to an increase in cluster quality and phone segmentation results compared to the base LDA. Compared to a recent vector quantised neural network approach that also learns 50 units, the extended LDA model performs better in phone segmentation but worse in mutual information.
In this paper, we ask whether vocal source features (pitch, shimmer, jitter, etc) can improve the performance of automatic sung speech recognition, arguing that conclusions previously drawn from spoken speech studies may not be valid in the sung speech domain. We first use a parallel singing/speaking corpus (NUS-48E) to illustrate differences in sung vs spoken voicing characteristics including pitch range, syllables duration, vibrato, jitter and shimmer. We then use this analysis to inform speech recognition experiments on the sung speech DSing corpus, using a state of the art acoustic model and augmenting conventional features with various voice source parameters. Experiments are run with three standard (increasingly large) training sets, DSing1 (15.1 hours), DSing3 (44.7 hours) and DSing30 (149.1 hours). Pitch combined with degree of voicing produces a significant decrease in WER from 38.1% to 36.7% when training with DSing1 however smaller decreases in WER observed when training with the larger more varied DSing3 and DSing30 sets were not seen to be statistically significant. Voicing quality characteristics did not improve recognition performance although analysis suggests that they do contribute to an improved discrimination between voiced/unvoiced phoneme pairs.
This paper describes the systems submitted to IWSLT 2021 by the Volctrans team. We participate in the offline speech translation and text-to-text simultaneous translation tracks. For offline speech translation, our best end-to-end model achieves 8.1 BLEU improvements over the benchmark on the MuST-C test set and is even approaching the results of a strong cascade solution. For text-to-text simultaneous translation, we explore the best practice to optimize the wait-k model. As a result, our final submitted systems exceed the benchmark at around 7 BLEU on the same latency regime. We will publish our code and model to facilitate both future research works and industrial applications.
Language modeling (LM) for automatic speech recognition (ASR) does not usually incorporate utterance level contextual information. For some domains like voice assistants, however, additional context, such as the time at which an utterance was spoken, provides a rich input signal. We introduce an attention mechanism for training neural speech recognition language models on both text and non-linguistic contextual data. When applied to a large de-identified dataset of utterances collected by a popular voice assistant platform, our method reduces perplexity by 7.0% relative over a standard LM that does not incorporate contextual information. When evaluated on utterances extracted from the long tail of the dataset, our method improves perplexity by 9.0% relative over a standard LM and by over 2.8% relative when compared to a state-of-the-art model for contextual LM.
Adversarial attacks have become a major threat for machine learning applications. There is a growing interest in studying these attacks in the audio domain, e.g, speech and speaker recognition; and find defenses against them. In this work, we focus on using representation learning to classify/detect attacks w.r.t. the attack algorithm, threat model or signal-to-adversarial-noise ratio. We found that common attacks in the literature can be classified with accuracies as high as 90%. Also, representations trained to classify attacks against speaker identification can be used also to classify attacks against speaker verification and speech recognition. We also tested an attack verification task, where we need to decide whether two speech utterances contain the same attack. We observed that our models did not generalize well to attack algorithms not included in the attack representation model training. Motivated by this, we evaluated an unknown attack detection task. We were able to detect unknown attacks with equal error rates of about 19%, which is promising.