Is there a way to allow deepgram to recognize "mhmm" as an affirmation as well as other sorts of filler words or speech sounds of this sort? It doesn't seem to be picking it up.

Hi there, Thank you for your message. Our team is currently out of the office. We operate Monday through Friday, from 9:00 AM to 8:00 PM Pacific Standard Time (PST). We’ll get back to you as soon as possible during our normal business hours. If your message is urgent, please mark it accordingly or include “URGENT” in the subject line, and we’ll do our best to respond promptly. Warm regards, Vapi Customer Support Team

To help Deepgram recognize "mhmm" and similar filler words, use the

keywords

parameter in your transcriber configuration to boost their detection. Here’s an example of how to add "mhmm" as a boosted keyword in your Vapi assistant setup:

json
"transcriber": {
  "provider": "deepgram",
  "model": "nova-2",
  "language": "en",
  "keywords": ["mhmm:5"]
}

This increases the likelihood that "mhmm" will be transcribed correctly. For best results, use the correct spelling and moderate the boost value as needed. (See the [Deepgram keyword boosting documentation](https://docs.vapi.ai/customization/custom-keywords) for more details.) Source: - [Deepgram Keywords and Keyterm Prompting](https://docs.vapi.ai/customization/custom-keywords)

Hi lotalop, Yes you can actually add filler tokens like "mhmm" to your assistant's Deepgram

keywords

(single words) or

keyterm

(multi‑word phrases). Example in the assistant transcriber config:

json
"transcriber": {
  "provider": "deepgram",
  "model": "nova-2",
  "language": "en",
  "keywords": ["mhmm:5", "hmm:3", "mmhmm:4"]
}

Notes and tips: - Use single tokens (no spaces) in `keywords`; use

keyterm

for phrases. See docs: https://docs.vapi.ai/customization/custom-keywords - Start with moderate integer intensifiers (e.g., 3–5) to boost detection without causing false positives. - Provide the exact spelling/casing variants you want recognized (e.g.,

mhmm

,

mmhmm

,

hmm

). - If you need realtime/dynamic injection of many tokens, that’s not currently supported — you must update the assistant config via the API. - If still not detected, try model variants (e.g., Nova vs Enhanced) or consider custom model training for robust filler-word handling.