Filler words

Filler words such as "um", "uh", "like", and "you know" are a natural part of spoken language. You can decide whether to include or exclude them from your transcript depending on what you're using the transcript for.

By default, Aldea removes filler words from the output to produce clean, readable text. However, you can preserve every utterance spoken in your verbatim by adding filler_words=true as a query parameter. This is useful for verbatim transcription or conversation analysis where hesitations carry meaning.

The following request shows how to include filler words in the output:

curl -X POST "https://api.aldea.ai/v1/listen?filler_words=true" \
  -H "Authorization: Bearer YOUR_ALDEA_API_KEY" \
  --data-binary @audio.wav

When to include filler words

Filler words carry meaningful signals in contexts such as:

• Conversation analysis and research: Linguists, UX researchers, and communication coaches study hesitation patterns to understand speaker confidence, cognitive load, and conversational dynamics. Removing fillers strips away the data they need.

• Legal and compliance transcription. Verbatim court transcripts, deposition records, and regulatory proceedings often require every spoken word to be documented. Omitting fillers could be considered an incomplete record.

• Sentiment and intent analysis. A customer who says "I, um, I guess the product is... fine" is communicating something different from one who says "The product is fine." Filler words and hesitation patterns can signal uncertainty, dissatisfaction, or discomfort that a cleaned-up transcript obscures.

• Speaker coaching and training. Sales teams, public speakers, and media trainers use filler word frequency as a measurable metric for improvement. Preserving fillers in training session transcripts makes that analysis possible.

When to exclude filler words

For most production use cases, the default behavior where filler words are removed produces the best result:

• Meeting notes and summaries: Most readers want the key take aways from a discussion, not a record of every word. Clean transcripts are faster to read and easier to summarize.

• Subtitles and captions: Screen space is limited, and some fillers may bring visual clutter without adding any value for the viewer.

• Content publishing: Podcast transcripts, interview articles, and blog posts read better without fillers. Published text follows written conventions, not spoken ones.

• Chatbot and voice assistant logs: When transcribing user commands or queries, fillers interfere with intent parsing. For example, the phrase "Um, set a timer for, uh, five minutes" could be harder to parse than "Set a timer for five minutes."

Next steps

• Smart formatting
• API reference