Why Accent Recognition Fails

Training Data Bias

STT models trained predominantly on:

• Standard American English (General American accent)
• Standard British English (Received Pronunciation)
• Studio-quality recordings with perfect audio conditions
• Native speakers with clear articulation

Result: Poor performance on non-standard accents, regional dialects, and real-world audio conditions.

Phonetic Variation Challenges

Same word, different pronunciations:

• "Water" pronounced with hard 't' (British) vs soft 'd' sound (American)
• "Schedule" with 'sh' sound (American) vs 'sk' sound (British)
• Regional vowel shifts changing word recognition patterns
• Consonant cluster reduction in some dialects

Missing Representation

Underrepresented accents in training data:

• African American Vernacular English (AAVE)
• Scottish, Irish, Welsh, and other UK regional accents
• Indian English variations
• Southeast Asian English accents
• Southern American English dialects
• Caribbean English varieties