Filler Words

Filler Words can transcribe disfluencies in your audio, like "uh" and "um".

Deepgram's Filler Words feature can transcribe disfluencies in your audio, like "uh" and "um".

Enable Feature

To enable Filler Words, when you call Deepgram’s API, set tier=nova&model=general&filler_words=true in the query string. This will transcribe filler words spoken in your audio into the body of the transcript.

ℹ️

Currently, Filler Words are only available for Deepgram's English Nova general model.

Deepgram is capable of transcribing the following filler words:

  • uh
  • um
  • mmhm
  • mm-mm
  • uh-uh
  • uh-huh
  • nuh-uh

These words will always be transcribed with the spelling listed above, regardless of their spoken duration (i.e., Deepgram will never transcribe "uhhhh" instead of "uh").

When filler_words=false or the parameter is not set, the two most common fillers, "uh" and "um", are stripped out of the transcript to improve readability.

curl \
  --request POST \
  --header 'Authorization: Token YOUR_DEEPGRAM_API_KEY' \
  --header 'Content-Type: audio/wav' \
  --data-binary @youraudio.wav \
  --url 'https://api.deepgram.com/v1/listen?tier=nova&model=general&filler_words=true'

:eyes: Replace YOUR_DEEPGRAM_API_KEY with your Deepgram API Key.

Results

Once applied, results will appear in the transcript.

TruthWith Filler WordsWithout Filler Words
uh-huh or you'd want something where uh so let's say you're trying to fine-tune a model to something very specific um so it's not as uh cut and dry as a more general taskuh-huh or you'd want something where uh so let's say you're trying to fine-tune a model to something very specific um so it's not as uh cut and dry as a more general taskuh-huh or you'd want something where so let's say you're trying to fine-tune a model to something very specific so it's not as cut and dry as a more general task

Use Cases

Some examples of use cases for Filler Words include:

  • Customers who need to collect analytics on the number of filler words spoken for coaching purposes.
  • Customers who need to remove filler words from the audio based on where they appear in transcripts.