Special Issue on Automatic Speech Recognition

With the rapid development of artificial intelligence and deep learning technology, automatic speech recognition technology is experiencing new vitality. However, there are still technological barriers to flexible solutions and user satisfaction in this field. This is due to several factors, such as sensitivity to the environment (background noise) or the weak representation of grammatical and semantic knowledge. There are actually many factors affecting speech realization: regional, sociolinguistic, environmental or personal. These create a wide range of speech variations that may not be correctly recognized and modeled.

This Special Issue aimed to collect and present all breakthrough research on speech segmentation and phoneme detection, speech recognition with noised speech, speech translation, classification of emotions in speech, and multimodal speech recognition with video or physiological signals.

A total of nine papers (seven research papers and two review papers) in different fields of automatic speech recognition, including speech recognition of different languages, speech enhancement, and speech emotion recognition, are presented in this Special Issue. Bhardwaj et al. [1] reported that children’s speech recognition (SR) is a challenging task due to the large variations in the articulatory, acoustic, physical, and linguistic characteristics of children’s speech compared to adult speech. Dhouib et al. [2] reported existing speech recognition studies related to Arabic ASR. Ali et al. [3] proposed a precise speech recognition system to overcome the issues of accents and local differences that affect an automatic speech recognition (ASR) system’s performance while analyzing speech signals. Kłosowski et al. [4] presented a rule-based grapheme-to-phoneme conversion method and algorithm for Polish. Song et al. [5] proposed a fully convolutional neural network based on recursive recurrent convolution for monaural speech enhancement in the time domain. Chen et al. [6] proposed a model utilizing only the fundamental frequency from electroglottograph (EGG) signals and cross-modal emotion distillation (CMED) to train the EGG-based speech emotion recognition (SER) model. Hossain et al. [7] created a dataset containing 30 h of Bangla speech of seven regional Bangla dialects with the goal of detecting synthesized Bangla speech and categorizing it. Dua et al. [8] developed a speech-to-text recognition system to recognize the tonal speech signals of Gurbani hymns using a CNN. Chen et al. [9] proposed a method for extracting features from the EGG of target speakers and separate target speakers from mixtures of different speakers in a noisy environment without clean speech.

Although submissions for this Special Issue are now closed, more in-depth research in the field of automatic speech recognition continues to address the challenges we continue to face today, such as multilingual automatic translation, automatic emotion recognition, and multimodal human–computer interactions.