Providers and ModelsDeepgram

Deepgram Nova 3

Deepgram Nova 3
Deepgram · Transcription
POST /v1/audio/transcriptions

Speech-to-text transcription using the Nova-3 model with multi-language support and advanced customizable settings for production workloads.

At a glance

FieldValue
Model iddeepgram-nova-3
Input modalitiesAudio
Output modalitiesText
Context window-
Weight precision-
Featurestranscription, speech_to_text
Native inferenceNo
NewNo
Supported endpointsPOST /v1/audio/transcriptions

Pricing

ChargeSpecRate
Transcriptionper minute of audio$0.014

Example request

$curl https://api.empiriolabs.ai/v1/audio/transcriptions \
>  -H 'Authorization: Bearer $EMPIRIOLABS_API_KEY' \
>  -F model=deepgram-nova-3 \
>  -F file=@meeting.mp3

Parameters

ParameterTypeRequiredDefaultDescription
filestringno-Audio file (mp3, wav, m4a, ogg, etc.) uploaded as multipart/form-data.
file_urlstringno-URL of the audio to transcribe. Mutually exclusive with file.
smart_formatbooleannotrueApply smart formatting (capitalize, punctuate, format numbers/dates).
profanity_filterbooleannofalseMask profane words in the transcript with ****.
numeralsbooleannofalseConvert spoken numbers to digits (e.g. “five” → “5”).
measurementsbooleannofalseConvert measurement words to abbreviations (e.g. “kilograms” → “kg”).
languageenumno"auto"ISO 639-1 language code. Auto-detect if omitted. · Allowed: auto, en, es, fr, de, it, pt, ja, zh, hi, ru, nl
redact_piibooleannofalseMask personally-identifiable information in the transcript.
redact_pcibooleannofalseMask payment card numbers in the transcript.
keytermstringno-Comma-separated key terms to boost recognition accuracy for (jargon, names).
diarizebooleannofalseIdentify different speakers and tag each utterance with a speaker label.
paragraphsbooleannotrueGroup output into paragraphs based on pauses and speaker turns.
dictationbooleannofalseInterpret commands like “period”, “comma”, “new line” as punctuation/formatting.
punctuatebooleannotrueAdd punctuation to the transcript.
utterancesbooleannofalseReturn each utterance separately with timing and confidence.
generate_pdfbooleannofalseWhen true, Deepgram returns a PDF transcript alongside the JSON.

Notes

Auto-detected languages

English, Spanish, French, German, Italian, Portuguese, Japanese, Chinese, Hindi, Russian, Dutch.

Custom Key Terms

  • English only, ≤500 tokens total
  • Improves recognition of domain-specific vocabulary

Feature interactions

  • Diarization auto-enables utterances
  • Smart format and paragraphs auto-enable punctuation
  • PCI/PII redaction and profanity filter are off by default

Machine-readable schema: GET https://api.empiriolabs.ai/v1/models/deepgram-nova-3.