wanghaisheng Voice Interaction Design
Design voice interfaces that work for the full range of human speech — including accents, speech disabilities, non-native speakers, and people in noisy or quiet environments.
Who This Is For
- People with motor disabilities who use voice as primary input
- People who stutter, have dysarthria, or other speech differences
- Non-native speakers with varied accents
- People in environments where typing is impractical
- Anyone who prefers voice to typing for certain tasks
Core Principles
Voice Should Never Be the Only Option
- Every voice interaction must have a text/touch/keyboard alternative
- Voice is an accelerator, not a gatekeeper
- Don't require voice for identity verification or critical actions unless an alternative exists
Design for Speech Variation
- Support varied pacing — don't cut off slow speakers
- Allow generous silence before timing out
- Don't penalise repetition, filler words, or self-correction
- Support multiple phrasings for the same intent ("go back", "previous page", "take me back", "undo")
Feedback Must Be Clear
- Confirm what the system heard (visual transcript)
- Make it easy to correct misrecognition
- Show when the system is listening vs. processing vs. waiting
- Never execute a destructive action on voice alone without confirmation
Design Patterns
Flexible Recognition
- Accept multiple ways to say the same command
- Don't require exact phrasing — intent matters more than syntax
- Support "did you mean?" clarification for ambiguous input
- Allow users to spell out words the system doesn't recognise
Graceful Failure
- When recognition fails: show what was heard and offer correction
- Never respond with just "I didn't understand" — offer alternatives
- Provide a "type instead" option at every failure point
- After repeated failures: proactively suggest switching to text input
Privacy and Control
- Clear visual/audio indicator when microphone is active
- Easy one-action mute/stop listening
- Don't record or transmit audio without explicit consent
- Allow users to review and delete voice data
Assessment Questions
- Can every voice-activated feature also be completed without voice?
- Does the system handle varied speech patterns without frustration?
- Is there clear feedback showing what the system heard?
- Can users easily correct misrecognition?
- Is it obvious when the system is listening?