Can AI Speak Like Us? : An Analysis of Accent Replication in Speech Synthesis

This study investigates the extent to which contemporary AI-based voice-cloning technology can replicate the phonetic characteristics of regional English accents. Combining principles from forensic phonetic analysis with sociolinguistic insights from World Englishes, four varieties of English: General American English (GAE), Scottish Standard English (SSE), Nigerian English (NigE), and Spanish English (SpE), are evaluated by comparing natural speech recordings with synthetic clones generated using the ElevenLabs platform. The aim is to determine whether cloned voices preserve both accentual identity and speaker-specific features, and to assess the implications for linguistic diversity, authenticity, and the emerging challenges posed by synthetic voice technologies. The results of the forensic phonetic analysis reveal a consistent pattern of prosodic simplification in synthetic speech. While the synthetic outputs broadly approximate the general characteristics of each accent, they fail to reproduce several accent-specific features essential for forensic or identity-based analyses. The findings highlight that despite advances in speech synthesis, accent fidelity and speaker individuality remain limited, with important consequences for voice-based identity representation, linguistic equity in AI systems, and the potential forensic misuse of synthetic voices. By situating the acoustic evidence within a broader sociolinguistic and ethical framework, this study demonstrates that current voice-cloning technologies risk reinforcing standard-accent biases and smoothing over the linguistic diversity represented in global Englishes. Further interdisciplinary work is necessary to improve model transparency, phonetic realism, and ethical safeguards as synthetic speech becomes increasingly integrated into communication technologies and public life.