Unveiling Biases while Embracing Sustainability: Assessing the Dual Challenges of Automatic Speech Recognition Systems

In this paper, we present a bias and sustainability focused investigation of Automatic Speech Recognition (ASR) systems, namely Whisper and Massively Multilingual Speech (MMS), which have achieved state-of-the-art (SOTA) performances. Despite their improved performance in controlled settings, there remains a critical gap in understanding their efficacy and equity in real-world scenarios. We analyze ASR biases w.r.t. gender, accent, and age group, as well as their effect on downstream tasks. In addition, we examine the environmental impact of ASR systems, scrutinizing the use of large acoustic models on carbon emission and energy consumption. We also provide insights into our empirical analyses, offering a valuable contribution to the claims surrounding bias and sustainability in ASR systems.

Unsupervised Rhythm and Voice Conversion of Dysarthric to Healthy Speech for ASR

Automatic speech recognition (ASR) systems are well known to perform poorly on dysarthric speech. Previous works have addressed this by speaking rate modification to reduce the mismatch with typical speech. Unfortunately, these approaches rely on transcribed speech data to estimate speaking rates and phoneme durations, which might not be available for unseen speakers. Therefore, we combine unsupervised rhythm and voice conversion methods based on self-supervised speech representations to map dysarthric to typical speech. We evaluate the outputs with a large ASR model pre-trained on healthy speech without further fine-tuning and find that the proposed rhythm conversion especially improves performance for speakers of the Torgo corpus with more severe cases of dysarthria. Code and audio samples are available at https://idiap.github.io/RnV .

SSL-TTS: Leveraging Self-Supervised Embeddings and kNN Retrieval for Zero-Shot Multi-speaker TTS

While recent zero-shot multispeaker text-to-speech (TTS) models achieve impressive results, they typically rely on extensive transcribed speech datasets from numerous speakers and intricate training pipelines. Meanwhile, self-supervised learning (SSL) speech features have emerged as effective intermediate representations for TTS. It was also observed that SSL features from different speakers that are linearly close share phonetic information while maintaining individual speaker identity, which enables straight-forward and robust voice cloning. In this study, we introduce SSL-TTS, a lightweight and efficient zero-shot TTS framework trained on transcribed speech from a single speaker. SSL-TTS leverages SSL features and retrieval methods for simple and robust zero-shot multi-speaker synthesis. Objective and subjective evaluations show that our approach achieves performance comparable to state-of-the-art models that require significantly larger training datasets. The low training data requirements mean that SSL-TTS is well suited for the development of multi-speaker TTS systems for low-resource domains and languages. We also introduce an interpolation parameter which enables fine control over the output speech by blending voices. Demo samples are available at https://idiap.github.io/ssl-tts

What Does it Take to Generalize SER Model Across Datasets? A Comprehensive Benchmark

Speech emotion recognition (SER) is essential for enhancing human-computer interaction in speech-based applications. Despite improvements in specific emotional datasets, there is still a research gap in SER's capability to generalize across real-world situations. In this paper, we investigate approaches to generalize the SER system across different emotion datasets. In particular, we incorporate 11 emotional speech datasets and illustrate a comprehensive benchmark on the SER task. We also address the challenge of imbalanced data distribution using over-sampling methods when combining SER datasets for training. Furthermore, we explore various evaluation protocols for adeptness in the generalization of SER. Building on this, we explore the potential of Whisper for SER, emphasizing the importance of thorough evaluation. Our approach is designed to advance SER technology by integrating speaker-independent methods.

The Balancing Act: Unmasking and Alleviating ASR Biases in Portuguese

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.

Training Convolutional Neural Networks with the Forward-Forward algorithm

The recent successes in analyzing images with deep neural networks are almost exclusively achieved with Convolutional Neural Networks (CNNs). The training of these CNNs, and in fact of all deep neural network architectures, uses the backpropagation algorithm where the output of the network is compared with the desired result and the difference is then used to tune the weights of the network towards the desired outcome. In a 2022 preprint, Geoffrey Hinton suggested an alternative way of training which passes the desired results together with the images at the input of the network. This so called Forward Forward (FF) algorithm has up to now only been used in fully connected networks. In this paper, we show how the FF paradigm can be extended to CNNs. Our FF-trained CNN, featuring a novel spatially-extended labeling technique, achieves a classification accuracy of 99.16% on the MNIST hand-written digits dataset. We show how different hyperparameters affect the performance of the proposed algorithm and compare the results with CNN trained with the standard backpropagation approach. Furthermore, we use Class Activation Maps to investigate which type of features are learnt by the FF algorithm.

ArTST: Arabic Text and Speech Transformer

We present ArTST, a pre-trained Arabic text and speech transformer for supporting open-source speech technologies for the Arabic language. The model architecture follows the unified-modal framework, SpeechT5, that was recently released for English, and is focused on Modern Standard Arabic (MSA), with plans to extend the model for dialectal and code-switched Arabic in future editions. We pre-trained the model from scratch on MSA speech and text data, and fine-tuned it for the following tasks: Automatic Speech Recognition (ASR), Text-To-Speech synthesis (TTS), and spoken dialect identification. In our experiments comparing ArTST with SpeechT5, as well as with previously reported results in these tasks, ArTST performs on a par with or exceeding the current state-of-the-art in all three tasks. Moreover, we find that our pre-training is conducive for generalization, which is particularly evident in the low-resource TTS task. The pre-trained model as well as the fine-tuned ASR and TTS models are released for research use.

Yet Another Model for Arabic Dialect Identification

In this paper, we describe a spoken Arabic dialect identification (ADI) model for Arabic that consistently outperforms previously published results on two benchmark datasets: ADI-5 and ADI-17. We explore two architectural variations: ResNet and ECAPA-TDNN, coupled with two types of acoustic features: MFCCs and features exratected from the pre-trained self-supervised model UniSpeech-SAT Large, as well as a fusion of all four variants. We find that individually, ECAPA-TDNN network outperforms ResNet, and models with UniSpeech-SAT features outperform models with MFCCs by a large margin. Furthermore, a fusion of all four variants consistently outperforms individual models. Our best models outperform previously reported results on both datasets, with accuracies of 84.7% and 96.9% on ADI-5 and ADI-17, respectively.

Adapting the adapters for code-switching in multilingual ASR

Recently, large pre-trained multilingual speech models have shown potential in scaling Automatic Speech Recognition (ASR) to many low-resource languages. Some of these models employ language adapters in their formulation, which helps to improve monolingual performance and avoids some of the drawbacks of multi-lingual modeling on resource-rich languages. However, this formulation restricts the usability of these models on code-switched speech, where two languages are mixed together in the same utterance. In this work, we propose ways to effectively fine-tune such models on code-switched speech, by assimilating information from both language adapters at each language adaptation point in the network. We also model code-switching as a sequence of latent binary sequences that can be used to guide the flow of information from each language adapter at the frame level. The proposed approaches are evaluated on three code-switched datasets encompassing Arabic, Mandarin, and Hindi languages paired with English, showing consistent improvements in code-switching performance with at least 10\% absolute reduction in CER across all test sets.

Self-supervised learning with diffusion-based multichannel speech enhancement for speaker verification under noisy conditions

The paper introduces Diff-Filter, a multichannel speech enhancement approach based on the diffusion probabilistic model, for improving speaker verification performance under noisy and reverberant conditions. It also presents a new two-step training procedure that takes the benefit of self-supervised learning. In the first stage, the Diff-Filter is trained by conducting timedomain speech filtering using a scoring-based diffusion model. In the second stage, the Diff-Filter is jointly optimized with a pre-trained ECAPA-TDNN speaker verification model under a self-supervised learning framework. We present a novel loss based on equal error rate. This loss is used to conduct selfsupervised learning on a dataset that is not labelled in terms of speakers. The proposed approach is evaluated on MultiSV, a multichannel speaker verification dataset, and shows significant improvements in performance under noisy multichannel conditions.