Remove Filler Words from Dictation Automatically — Um, Uh, No More

Natural speech is full of filler words — "um", "uh", "erm", "er", "ah", "hmm." When you're dictating a document or email, these fillers clutter the output. Whisperer removes them automatically.

Enabling Filler Word Removal#

Filler word removal is off by default (to preserve exact transcription when needed). To enable it:

1. Open Whisperer Settings
2. Find "Filler Word Removal" under Text Processing
3. Toggle it on

That's it. From now on, all filler words are stripped from your dictation output.

What Gets Removed#

Whisperer removes these filler words using word-boundary matching:

• um
• uh
• erm
• er
• ah
• hmm

The Text Processing Pipeline#

Filler word removal is one step in Whisperer's full text processing pipeline. Every transcription passes through these stages in order:

1. Transcription — Whisper/Parakeet/Apple Speech produces raw text
2. Dictionary corrections — SymSpell + phonetic + exact matching fixes technical terms
3. Filler word removal — Strips um, uh, erm, er, ah, hmm (optional)
4. List formatting — Converts spoken lists to formatted bullet points (optional)
5. AI post-processing — Rewrite, translate, format, summarize (optional)
6. Trailing space — Appends a space so the cursor is ready for the next word (optional)
7. Text injection — Final text inserted into the focused field

Each step is independently toggleable. You can use any combination that suits your workflow.

List Formatting#

Whisperer includes a deterministic list detection engine that converts spoken enumerations to formatted lists — no AI required.

It detects multiple marker types:

• Spoken ordinals: "first", "second", "third"
• Cardinals: "one", "two", "three"
• Digits: "1", "2", "3"
• "Number X" phrases: "number one", "number two"
• Bullet triggers: "bullet point", "dash"

The engine uses 5 detection strategies and includes extensive false-positive prevention with 60+ blocked preceding words — phrases like "have two coffees" or "about three hours" are correctly identified as regular speech, not list markers.

Optional AI fallback: When the deterministic engine doesn't find a list, an optional LLM fallback can analyze the text for implicit list structure.

When to Use What#

Getting Started#

1. Download Whisperer
2. Enable Filler Word Removal in Settings
3. Optionally enable List Formatting for structured dictation
4. Explore AI post-processing for even more text cleanup