Despite advancements in speech recognition, accented speech remains challenging. While previous approaches have focused on modeling techniques or creating accented speech datasets, gathering sufficient data for the multitude of accents, particularly in the African context, remains impractical due to their sheer diversity and associated budget constraints. To address these challenges, we propose AccentFold, a method that exploits spatial relationships between learned accent embeddings to improve downstream Automatic Speech Recognition (ASR). Our exploratory analysis of speech embeddings representing 100+ African accents reveals interesting spatial accent relationships highlighting geographic and genealogical similarities, capturing consistent phonological, and morphological regularities, all learned empirically from speech. Furthermore, we discover accent relationships previously uncharacterized by the Ethnologue. Through empirical evaluation, we demonstrate the effectiveness of AccentFold by showing that, for out-of-distribution (OOD) accents, sampling accent subsets for training based on AccentFold information outperforms strong baselines a relative WER improvement of 4.6%. AccentFold presents a promising approach for improving ASR performance on accented speech, particularly in the context of African accents, where data scarcity and budget constraints pose significant challenges. Our findings emphasize the potential of leveraging linguistic relationships to improve zero-shot ASR adaptation to target accents.
In this paper, we propose an efficient and accurate streaming speech recognition model based on the FastConformer architecture. We adapted the FastConformer architecture for streaming applications through: (1) constraining both the look-ahead and past contexts in the encoder, and (2) introducing an activation caching mechanism to enable the non-autoregressive encoder to operate autoregressively during inference. The proposed model is thoughtfully designed in a way to eliminate the accuracy disparity between the train and inference time which is common for many streaming models. Furthermore, our proposed encoder works with various decoder configurations including Connectionist Temporal Classification (CTC) and RNN-Transducer (RNNT) decoders. Additionally, we introduced a hybrid CTC/RNNT architecture which utilizes a shared encoder with both a CTC and RNNT decoder to boost the accuracy and save computation. We evaluate the proposed model on LibriSpeech dataset and a multi-domain large scale dataset and demonstrate that it can achieve better accuracy with lower latency and inference time compared to a conventional buffered streaming model baseline. We also showed that training a model with multiple latencies can achieve better accuracy than single latency models while it enables us to support multiple latencies with a single model. Our experiments also showed the hybrid architecture would not only speedup the convergence of the CTC decoder but also improves the accuracy of streaming models compared to single decoder models.
Deploying end-to-end speech recognition models with limited computing resources remains challenging, despite their impressive performance. Given the gradual increase in model size and the wide range of model applications, selectively executing model components for different inputs to improve the inference efficiency is of great interest. In this paper, we propose a dynamic layer-skipping method that leverages the CTC blank output from intermediate layers to trigger the skipping of the last few encoder layers for frames with high blank probabilities. Furthermore, we factorize the CTC output distribution and perform knowledge distillation on intermediate layers to reduce computation and improve recognition accuracy. Experimental results show that by utilizing the CTC blank, the encoder layer depth can be adjusted dynamically, resulting in 29% acceleration of the CTC model inference with minor performance degradation.
We propose a novel neural network architecture based on conformer transducer that adds contextual information flow to the ASR systems. Our method improves the accuracy of recognizing uncommon words while not harming the word error rate of regular words. We explore the uncommon words accuracy improvement when we use the new model and/or shallow fusion with context language model. We found that combination of both provides cumulative gain in uncommon words recognition accuracy.
This paper proposes a novel, resource-efficient approach to Visual Speech Recognition (VSR) leveraging speech representations produced by any trained Automatic Speech Recognition (ASR) model. Moving away from the resource-intensive trends prevalent in recent literature, our method distills knowledge from a trained Conformer-based ASR model, achieving competitive performance on standard VSR benchmarks with significantly less resource utilization. Using unlabeled audio-visual data only, our baseline model achieves a word error rate (WER) of 47.4% and 54.7% on the LRS2 and LRS3 test benchmarks, respectively. After fine-tuning the model with limited labeled data, the word error rate reduces to 35% (LRS2) and 45.7% (LRS3). Our model can be trained on a single consumer-grade GPU within a few days and is capable of performing real-time end-to-end VSR on dated hardware, suggesting a path towards more accessible and resource-efficient VSR methodologies.
Cued Speech (CS) is a pure visual coding method used by hearing-impaired people that combines lip reading with several specific hand shapes to make the spoken language visible. Automatic CS recognition (ACSR) seeks to transcribe visual cues of speech into text, which can help hearing-impaired people to communicate effectively. The visual information of CS contains lip reading and hand cueing, thus the fusion of them plays an important role in ACSR. However, most previous fusion methods struggle to capture the global dependency present in long sequence inputs of multi-modal CS data. As a result, these methods generally fail to learn the effective cross-modal relationships that contribute to the fusion. Recently, attention-based transformers have been a prevalent idea for capturing the global dependency over the long sequence in multi-modal fusion, but existing multi-modal fusion transformers suffer from both poor recognition accuracy and inefficient computation for the ACSR task. To address these problems, we develop a novel computation and parameter efficient multi-modal fusion transformer by proposing a novel Token-Importance-Aware Attention mechanism (TIAA), where a token utilization rate (TUR) is formulated to select the important tokens from the multi-modal streams. More precisely, TIAA firstly models the modality-specific fine-grained temporal dependencies over all tokens of each modality, and then learns the efficient cross-modal interaction for the modality-shared coarse-grained temporal dependencies over the important tokens of different modalities. Besides, a light-weight gated hidden projection is designed to control the feature flows of TIAA. The resulting model, named Economical Cued Speech Fusion Transformer (EcoCued), achieves state-of-the-art performance on all existing CS datasets, compared with existing transformer-based fusion methods and ACSR fusion methods.
In the field of spoken language understanding, systems like Whisper and Multilingual Massive Speech (MMS) have shown state-of-the-art performances. This study is dedicated to a comprehensive exploration of the Whisper and MMS systems, with a focus on assessing biases in automatic speech recognition (ASR) inherent to casual conversation speech specific to the Portuguese language. Our investigation encompasses various categories, including gender, age, skin tone color, and geo-location. Alongside traditional ASR evaluation metrics such as Word Error Rate (WER), we have incorporated p-value statistical significance for gender bias analysis. Furthermore, we extensively examine the impact of data distribution and empirically show that oversampling techniques alleviate such stereotypical biases. This research represents a pioneering effort in quantifying biases in the Portuguese language context through the application of MMS and Whisper, contributing to a better understanding of ASR systems' performance in multilingual settings.
Lipreading involves using visual data to recognize spoken words by analyzing the movements of the lips and surrounding area. It is a hot research topic with many potential applications, such as human-machine interaction and enhancing audio speech recognition. Recent deep-learning based works aim to integrate visual features extracted from the mouth region with landmark points on the lip contours. However, employing a simple combination method such as concatenation may not be the most effective approach to get the optimal feature vector. To address this challenge, firstly, we propose a cross-attention fusion-based approach for large lexicon Arabic vocabulary to predict spoken words in videos. Our method leverages the power of cross-attention networks to efficiently integrate visual and geometric features computed on the mouth region. Secondly, we introduce the first large-scale Lip Reading in the Wild for Arabic (LRW-AR) dataset containing 20,000 videos for 100-word classes, uttered by 36 speakers. The experimental results obtained on LRW-AR and ArabicVisual databases showed the effectiveness and robustness of the proposed approach in recognizing Arabic words. Our work provides insights into the feasibility and effectiveness of applying lipreading techniques to the Arabic language, opening doors for further research in this field. Link to the project page: https://crns-smartvision.github.io/lrwar
Pre-trained Models (PTMs) have facilitated substantial progress in the field of Speech Emotion Recognition (SER). SER is an area with applications ranging from HumanComputer Interaction to Healthcare. Recent studies have leveraged various PTM representations as input features for downstream models for SER. PTM specifically pre-trained for paralinguistic tasks have obtained state-of-the-art (SOTA) performance for SER. However, such PTM haven't been evaluated for SER in multilingual settings and experimented only with English. So, we fill this gap, by performing a comprehensive comparative study of five PTMs (TRILLsson, wav2vec2, XLS-R, x-vector, Whisper) for assessing the effectiveness of paralingual PTM (TRILLsson) for SER across multiple languages. Representations from TRILLsson achieved the best performance among all the PTMs. This demonstrates that TRILLsson is able to effectively capture the various paralinguistic features from speech data for better SER. We also show that downstream models using TRILLsson representations achieve SOTA performance in terms of accuracy across various multi-lingual datasets.
While recent work shows promising results in expanding the capabilities of large language models (LLM) to directly understand and synthesize speech, an LLM-based strategy for modeling spoken dialogs remains elusive and calls for further investigation. This work proposes an extensive speech-text LLM framework, named the Unified Spoken Dialog Model (USDM), to generate coherent spoken responses with organic prosodic features relevant to the given input speech without relying on automatic speech recognition (ASR) or text-to-speech (TTS) solutions. Our approach employs a multi-step speech-text inference scheme that leverages chain-of-reasoning capabilities exhibited by the underlying LLM. We also propose a generalized speech-text pretraining scheme that helps with capturing cross-modal semantics. Automatic and human evaluations show that the proposed approach is effective in generating natural-sounding spoken responses, outperforming both prior and cascaded baselines. Detailed comparative studies reveal that, despite the cascaded approach being stronger in individual components, the joint speech-text modeling improves robustness against recognition errors and speech quality. Demo is available at https://unifiedsdm.github.io.