Blind users rely on fast, predictable text-to-speech voices, but the dominant Eloquence voice hasn't been updated since 2003. Modern AI systems like Supertonic and Kitten TTS promise speed but introduce new problems for accessibility.

These AI models require massive dependency bloat—over 100 Python packages for Kitten TTS—slowing screen readers and creating security risks. They also sacrifice accuracy, skipping words and misreading numbers, which is unacceptable for users who navigate text at 800+ words per minute.

The fundamental issue is control. Traditional voices let users adjust pitch, speed, and breathiness in real-time. AI voices inherit fixed characteristics from training data, offering only basic speed controls. This breaks critical workflows for screen reader users who depend on immediate, customizable speech output.