Disfluency removal
Removing filler words like 'um' and 'uh' from speech for clearer audio.
Disfluency removal is a process that cleans up spoken audio by getting rid of sounds and words people naturally use when they pause or hesitate. This includes filler words like 'um,' 'uh,' 'like,' and 'you know,' as well as false starts where someone begins a sentence and then restarts it.
Imagine listening to a podcast. If the speaker says 'I went to the store and, um, I bought some milk,' disfluency removal would clean it up to sound like 'I went to the store and I bought some milk.' This makes the audio easier to understand and sound more professional.
Disfluency removal is an automated process applied to speech audio or transcripts to eliminate non-linguistic utterances and false starts. This typically involves identifying and removing filler words (e.g., 'uh', 'um'), repetitions, and self-corrections. The goal is to produce a cleaner, more fluent output that is easier for humans or downstream AI systems to process.
Techniques often involve acoustic modeling to detect filler sounds and language modeling to predict and correct hesitations or restarts. Tradeoffs include potential loss of naturalness if overdone, or failure to remove subtle disfluencies. Advanced systems aim for a balance, preserving conversational flow while enhancing clarity.
This term is appearing more frequently due to advancements in AI-powered transcription services. Companies like Google Gemini are integrating disfluency removal into their APIs, making it easier for developers to create cleaner audio outputs for applications like voice assistants and meeting summarizers.
- Gemini API Cookbook: Add Smart transcription section to Gemini Transcribe quickstart (#1349)Google · Gemini API Cookbook ·
- Google Gemini
- Google Gemini's API offers smart transcription features that include disfluency removal, enhancing the clarity of transcribed audio.
- Noise reduction
- Noise reduction focuses on removing background sounds, not spee
- Punctuation restoration
- Punctuation restoration adds grammatical marks to text,
- Speaker diarization
- Speaker diarization identifies who is speaking when, not wh
The concept of removing speech disfluencies has roots in linguistics and speech processing research, aiming to create more understandable speech. Automated methods have evolved with advances in machine learning and AI.
Record yourself speaking a sentence with filler words, like 'I think, uh, we should go to the, um, park.' Then, use a transcription tool that offers disfluency removal, such as those available through Google Gemini's API documentation, to see how it cleans up your speech.
The Playbook: reusable how-tos, prompts, configs and tips pulled out of the news.
- PromptRole-play a student struggling with math for AI tutor simulationSimulate a student's math struggles and uncertainty for AI tutor evaluation. · 5 min
- TemplateReusable workflow templates for healthcare teamsTeams can create and share templates for common workflows like documentation or prior authorizations. · 15 min
- TemplateTemplate for building an AI tool tailored to your workA structured approach to defining and building a custom AI tool for specific business needs. · 15 min
- TemplateEvals CSV prompt list for agent skillsA CSV file template for listing prompts used in evaluating agent skills. · 20 min
- HeuristicDouble coding agent task completion with docs linksProviding coding agents with a documentation link significantly increases their task completion rate. · 2 min
Questions people ask
- What are examples of disfluencies?
- Common disfluencies include filler words like 'um,' 'uh,' 'like,' and 'you know,' as well as repetitions of words or sounds and false starts where a speaker begins a phrase and then changes their mind.
- Why is disfluency removal useful?
- It makes spoken content easier to understand and more professional sounding, which is helpful for automated summaries, voice assistants, and anyone listening to recorded speech.
First seen on Subvolts Aug 26, 2026. Updated Sep 15, 2026. Written by an AI model from the items above and reviewed against them weekly; report an error.