Real-Time Voice: A Comprehensive Survey of Automatic Speech Recognition and Transcription

Human-computer interaction has been transformed by Automatic Speech Recognition (ASR), opening the door to uses including voice-activated devices, assistive technology, and real-time transcription. But it’s still difficult to get reliable results when dealing with different accents, loud surroundings, and domain-specific jargon. In this research, a high-performance ASR system utilizing a WaveNet-based architecture and sophisticated preprocessing methods such as SubSpectral Normalization and Mel-Frequency Cepstral Coefficients is presented. The suggested approach combines linguistic decoders with causal and dilated convolutions to improve contextual comprehension. Time-stretching and noise addition are two data augmentation techniques that guarantee resilience to real-world fluctuations. Results from experiments show notable gains in generalization, scalability, and transcription accuracy. By providing useful answers for a range of applications, including accessibility tools and virtual assistants, this work advances ASR technologies.