Remy Reference/models/Deepgram Nova-3

Deepgram Nova-3

Deepgram Nova-3, a transcription model from Deepgram.

transcriptionModelOverride: { model: "deepgram-nova-3" }

Deepgram·Released February 2025

Pricing

List price: $0.0043 / min of audio
Per second: $0.000072
Per minute: $0.0043
Per second: $0.000087
Per minute: $0.0052
Per second: $0.000022
Per minute: $0.0013
Per second: $0.000033
Per minute: $0.0020

Specifications

Released: February 2025
Model string: nova-3
Capabilities: RealtimeLow latency

Parameters

Model options, passed in transcriptionModelOverride.config.

Parameter	Type	Default	Range / options
`includeSpeakers` Label each utterance with a speaker index when multiple people talk.	Select	no	yesno
`language` The spoken language, as a code such as en or es.	Select	en	multienesfrdeit+8
`detectLanguage` Detect the spoken language instead of assuming English. Leave off for known-English audio.	Select	no	yesno
`keyterms` Names, products and terms the audio contains, so the model spells them right.	Text list	—	—
`redactPii` Replace names, phone numbers, addresses and other personal information in the transcript with a label of what was there.	Select	no	yesno
`includeWordTimings` Add per-word start/end times and confidence inside each transcript segment. Increases the size of the result considerably.	Select	no	yesno
`subtitles` Also return the transcript as a subtitle file, SRT or WebVTT, timed to the audio.	Select	—	srtvtt

Use this model

await mindstudio.transcribeAudio({
  audioUrl: "...",
  transcriptionModelOverride: {
    model: "deepgram-nova-3",
    config: {
      includeSpeakers: "no",
      language: "en",
    },
  },
});