Grammar/spell correction of transcribed text (cheap model) before auto-paste

[remziduzagac]

Hi

It would be really helpful—especially for non-native English speakers (and even for some native speakers)—to optionally run a lightweight grammar/spell correction step on the transcribed text before it’s copied/auto-pasted.

This would help with:

• Fixing basic grammar and punctuation
• Correcting common transcription mistakes (homophones, missing words, casing)
• Producing cleaner output for prompts/messages/docs without extra manual edits

Suggested approach

• Make it optional (flag/config), default off
• Use a simple/cheap model (or even rules-based) to keep latency/cost low
• Prefer a “light touch”: correct errors without changing the user’s wording or tone
• Ideally apply to both batch (Whisper) and streaming (Deepgram) modes if feasible

Behavior rules (proposed)

• Only do: punctuation, spelling, capitalization, minimal grammar
• Do not rewrite sentences for style, conciseness, or tone
• Preserve meaning strictly (no paraphrasing)
• Avoid expanding/altering technical terms, code, file paths, URLs, commands, etc.
• Optionally: allow a “strictness” setting (e.g., light vs medium)

Config ideas (example)

• post_process = none | grammar
• grammar_provider = <provider>
• grammar_model = <small/cheap model>
• Optional strictness = light | medium

Thanks!