Continuous speech recognition for Tamil language using a novel semantic verification integrated with the transformer model

Automatic speech recognition (ASR) is a technology that decodes and transcribes spoken language into text. Using a microphone to capture audio from a speaker, ASR systems process this input through algorithms or models to produce written output, which is typically in textual form. This technology is crucial for low-resource languages, which ensures efficient transcription in applications like transcription services, voice assistants, and language processing tools. It plays a pivotal role in facilitating effective communication and information access in languages with limited linguistic resources like Tamil. However, accurate transcription of spoken words into written text is challenging due to the variability in pronunciation, intonation, accent, and speed present in speech signals. Additionally, environmental factors such as background noise and speaker variations further complicate the task. In this work, we propose a hybrid model known as a “Convolutional Transformer with Semantic Verification” (CTSV) to enhance speaker recognition system for the Tamil language. Experimental validation is carried out on a Tamil speech corpus along with their corresponding transcriptions. The proposed model is evaluated on a test dataset under noisy and noiseless scenarios using metrics like word error rate (WER), character error rate (CER), and word information lost (WIL) which yield an error rate of 12.09%, 2.44%, and 19.5%, respectively. These findings demonstrate the model’s effectiveness when compared to baseline ASR models.