In this paper, we present a 170.83 hour Indian English spontaneous speech dataset. Lack of Indian English speech data is one of the major hindrances in developing robust speech systems which are adapted to the Indian speech style. Moreover this scarcity is even more for spontaneous speech. This corpus is crowd sourced over varied Indian nativities, genders and age groups. Traditional spontaneous speech collection strategies involve capturing of speech during interviewing or conversations. In this study, we use images as stimuli to induce spontaneity in speech. Transcripts for 23 hours is generated and validated which can serve as a spontaneous speech ASR benchmark. Quality of the corpus is validated with voice activity detection based segmentation, gender verification and image semantic correlation. Which determines a relationship between image stimulus and recorded speech using caption keywords derived from Image2Text model and high occurring words derived from whisper ASR generated transcripts.
Machine learning models for speech emotion recognition (SER) can be trained for different tasks and are usually evaluated on the basis of a few available datasets per task. Tasks could include arousal, valence, dominance, emotional categories, or tone of voice. Those models are mainly evaluated in terms of correlation or recall, and always show some errors in their predictions. The errors manifest themselves in model behaviour, which can be very different along different dimensions even if the same recall or correlation is achieved by the model. This paper investigates behavior of speech emotion recognition models with a testing framework which requires models to fulfill conditions in terms of correctness, fairness, and robustness.
Large-scale automatic speech translation systems today lack key features that help machine-mediated communication feel seamless when compared to human-to-human dialogue. In this work, we introduce a family of models that enable end-to-end expressive and multilingual translations in a streaming fashion. First, we contribute an improved version of the massively multilingual and multimodal SeamlessM4T model-SeamlessM4T v2. This newer model, incorporating an updated UnitY2 framework, was trained on more low-resource language data. SeamlessM4T v2 provides the foundation on which our next two models are initiated. SeamlessExpressive enables translation that preserves vocal styles and prosody. Compared to previous efforts in expressive speech research, our work addresses certain underexplored aspects of prosody, such as speech rate and pauses, while also preserving the style of one's voice. As for SeamlessStreaming, our model leverages the Efficient Monotonic Multihead Attention mechanism to generate low-latency target translations without waiting for complete source utterances. As the first of its kind, SeamlessStreaming enables simultaneous speech-to-speech/text translation for multiple source and target languages. To ensure that our models can be used safely and responsibly, we implemented the first known red-teaming effort for multimodal machine translation, a system for the detection and mitigation of added toxicity, a systematic evaluation of gender bias, and an inaudible localized watermarking mechanism designed to dampen the impact of deepfakes. Consequently, we bring major components from SeamlessExpressive and SeamlessStreaming together to form Seamless, the first publicly available system that unlocks expressive cross-lingual communication in real-time. The contributions to this work are publicly released and accessible at https://github.com/facebookresearch/seamless_communication
In the realm of automatic speech recognition (ASR), the quest for models that not only perform with high accuracy but also offer transparency in their decision-making processes is crucial. The potential of quality estimation (QE) metrics is introduced and evaluated as a novel tool to enhance explainable artificial intelligence (XAI) in ASR systems. Through experiments and analyses, the capabilities of the NoRefER (No Reference Error Rate) metric are explored in identifying word-level errors to aid post-editors in refining ASR hypotheses. The investigation also extends to the utility of NoRefER in the corpus-building process, demonstrating its effectiveness in augmenting datasets with insightful annotations. The diagnostic aspects of NoRefER are examined, revealing its ability to provide valuable insights into model behaviors and decision patterns. This has proven beneficial for prioritizing hypotheses in post-editing workflows and fine-tuning ASR models. The findings suggest that NoRefER is not merely a tool for error detection but also a comprehensive framework for enhancing ASR systems' transparency, efficiency, and effectiveness. To ensure the reproducibility of the results, all source codes of this study are made publicly available.
When performing tasks like automatic speech recognition or spoken language understanding for a given utterance, access to preceding text or audio provides contextual information can improve performance. Considering the recent advances in generative large language models (LLM), we hypothesize that an LLM could generate useful context information using the preceding text. With appropriate prompts, LLM could generate a prediction of the next sentence or abstractive text like titles or topics. In this paper, we study the use of LLM-generated context information and propose an approach to distill the generated information during fine-tuning of self-supervised speech models, which we refer to as generative context-aware fine-tuning. This approach allows the fine-tuned model to make improved predictions without access to the true surrounding segments or to the LLM at inference time, while requiring only a very small additional context module. We evaluate the proposed approach using the SLUE and Libri-light benchmarks for several downstream tasks: automatic speech recognition, named entity recognition, and sentiment analysis. The results show that generative context-aware fine-tuning outperforms a context injection fine-tuning approach that accesses the ground-truth previous text, and is competitive with a generative context injection fine-tuning approach that requires the LLM at inference time.
With increasingly more powerful compute capabilities and resources in today's devices, traditionally compute-intensive automatic speech recognition (ASR) has been moving from the cloud to devices to better protect user privacy. However, it is still challenging to implement on-device ASR on resource-constrained devices, such as smartphones, smart wearables, and other small home automation devices. In this paper, we propose a series of model architecture adaptions, neural network graph transformations, and numerical optimizations to fit an advanced Conformer based end-to-end streaming ASR system on resource-constrained devices without accuracy degradation. We achieve over 5.26 times faster than realtime (0.19 RTF) speech recognition on small wearables while minimizing energy consumption and achieving state-of-the-art accuracy. The proposed methods are widely applicable to other transformer-based server-free AI applications. In addition, we provide a complete theory on optimal pre-normalizers that numerically stabilize layer normalization in any Lp-norm using any floating point precision.
Language is a dynamic aspect of our culture that changes when expressed in different technologies/communities. Online social networks have enabled the diffusion and evolution of different dialects, including African American English (AAE). However, this increased usage is not without barriers. One particular barrier is how sentiment (Vader, TextBlob, and Flair) and toxicity (Google's Perspective and the open-source Detoxify) methods present biases towards utterances with AAE expressions. Consider Google's Perspective to understand bias. Here, an utterance such as ``All n*ggers deserve to die respectfully. The police murder us.'' it reaches a higher toxicity than ``African-Americans deserve to die respectfully. The police murder us.''. This score difference likely arises because the tool cannot understand the re-appropriation of the term ``n*gger''. One explanation for this bias is that AI models are trained on limited datasets, and using such a term in training data is more likely to appear in a toxic utterance. While this may be plausible, the tool will make mistakes regardless. Here, we study bias on two Web-based (YouTube and Twitter) datasets and two spoken English datasets. Our analysis shows how most models present biases towards AAE in most settings. We isolate the impact of AAE expression usage via linguistic control features from the Linguistic Inquiry and Word Count (LIWC) software, grammatical control features extracted via Part-of-Speech (PoS) tagging from Natural Language Processing (NLP) models, and the semantic of utterances by comparing sentence embeddings from recent language models. We present consistent results on how a heavy usage of AAE expressions may cause the speaker to be considered substantially more toxic, even when speaking about nearly the same subject. Our study complements similar analyses focusing on small datasets and/or one method only.
Event-related potentials (ERP) have been used to address a wide range of research questions in neuroscience and cognitive psychology including selective auditory attention. The recent progress in auditory attention decoding (AAD) methods is based on algorithms that find a relation between the audio envelope and the neurophysiological response. The most popular approach is based on the reconstruction of the audio envelope based on EEG signals. However, these methods are mainly based on the neurophysiological entrainment to physical attributes of the sensory stimulus and are generally limited by a long detection window. This study proposes a novel approach to auditory attention decoding by looking at higher-level cognitive responses to natural speech. To investigate if natural speech events elicit cognitive ERP components and how these components are affected by attention mechanisms, we designed a series of four experimental paradigms with increasing complexity: a word category oddball paradigm, a word category oddball paradigm with competing speakers, and competing speech streams with and without specific targets. We recorded the electroencephalogram (EEG) from 32 scalp electrodes and 12 in-ear electrodes (ear-EEG) from 24 participants. A cognitive ERP component, which we believe is related to the well-known P3b component, was observed at parietal electrode sites with a latency of approximately 620 ms. The component is statistically most significant for the simplest paradigm and gradually decreases in strength with increasing complexity of the paradigm. We also show that the component can be observed in the in-ear EEG signals by using spatial filtering. The cognitive component elicited by auditory attention may contribute to decoding auditory attention from electrophysiological recordings and its presence in the ear-EEG signals is promising for future applications within hearing aids.
We propose a multi-task learning (MTL) model for jointly performing three tasks that are commonly solved in a text-to-speech (TTS) front-end: text normalization (TN), part-of-speech (POS) tagging, and homograph disambiguation (HD). Our framework utilizes a tree-like structure with a trunk that learns shared representations, followed by separate task-specific heads. We further incorporate a pre-trained language model to utilize its built-in lexical and contextual knowledge, and study how to best use its embeddings so as to most effectively benefit our multi-task model. Through task-wise ablations, we show that our full model trained on all three tasks achieves the strongest overall performance compared to models trained on individual or sub-combinations of tasks, confirming the advantages of our MTL framework. Finally, we introduce a new HD dataset containing a balanced number of sentences in diverse contexts for a variety of homographs and their pronunciations. We demonstrate that incorporating this dataset into training significantly improves HD performance over only using a commonly used, but imbalanced, pre-existing dataset.
Transformer structures have demonstrated outstanding skills in the deep learning space recently, significantly increasing the accuracy of models across a variety of domains. Researchers have started to question whether such a sophisticated network structure is actually necessary and whether equally outstanding results can be reached with reduced inference cost due to its complicated network topology and high inference cost. In order to prove the Mixer's efficacy on three datasets Speech Commands, UrbanSound8k, and CASIA Chinese Sentiment Corpus this paper applies amore condensed version of the Mixer to an audio classification task and conducts comparative experiments with the Transformer-based Audio Spectrogram Transformer (AST)model. In addition, this paper conducts comparative experiments on the application of several activation functions in Mixer, namely GeLU, Mish, Swish and Acon-C. Further-more, the use of various activation functions in Mixer, including GeLU, Mish, Swish, and Acon-C, is compared in this research through comparison experiments. Additionally, some AST model flaws are highlighted, and the model suggested in this study is improved as a result. In conclusion, a model called the Audio Spectrogram Mixer, which is the first model for audio classification with Mixer, is suggested in this study and the model's future directions for improvement are examined.