The traditional speech transcription pipeline consists of two stages: speech acquisition and the conversion of spoken language into written text using Automatic Speech Recognition (ASR) systems. Despite substantial advances in recent years, ASR technologies continue to exhibit limitations, particularly in handling punctuation, homophones, and linguistic disfluencies. This article proposes an extension of the conventional pipeline by introducing a third stage – post-processing in which generative AI is used to enhance transcription quality. Using a large language model guided by a carefully crafted prompt, raw transcripts are refined for coherence, accuracy, and proper punctuation, while minimizing unintended alterations. An empirical evaluation was conducted on 224 audio samples, including both human and AI-generated voices, with variations in speech rate and three acoustic disturbances: recording distortions, train noise, and crowd noise. Results demonstrate substantial improvements: punctuation restoration, correction of homophone errors in over 43% of cases, and elimination of hesitations or repetitions in 74% of transcripts. Overall, 35% of outputs were subjectively rated as superior, with the most pronounced benefits observed in non-native speakers’ speech.