Claude APISpeech-to-TextNext.jsStreaming

Voice Polisher

A voice-to-text refinement tool that preserves the speaker’s intent while elevating clarity and grammar.

The Problem

Voice-to-text transcription is fast but messy. The output reads like someone talking, not writing. Existing grammar tools strip personality and flatten voice into generic corporate prose.

Creatives and professionals need transcription that sounds like them, just cleaner.

The Approach

Built a two-pass refinement pipeline: the first pass corrects grammar and structure while preserving voice markers (rhythm, word choice, emphasis). The second pass applies a user-configurable “formality dial” that adjusts tone without losing authenticity.

Streaming output lets users watch the refinement happen in real time.

Key Screens

96%
Voice preservation score
1.8s
Time to first token
40%
Fewer manual edits needed

What I Learned

The biggest insight was that ‘polishing’ isn’t about making text perfect—it’s about making it sound like the person intended. Personality is a feature, not a bug.

Tech Stack

Next.jsClaude APIWhisperVercel AI SDKTailwind CSS

Interested in working together?

Get in touch