Deepgram

Deepgram Nova models are fast cloud speech-to-text models with rich pre-recorded audio options: language detection, diarization, formatting, redaction, search, topic detection, sentiment, and keyword/keyterm prompting.

API documentation: Deepgram pre-recorded listen API

Models

Model ID	Notes
`deepgram/nova-3`	Latest Nova model. Supports `keyterm` prompting.
`deepgram/nova-2`	Previous Nova generation. Uses `keywords` prompting.

Select A Model

bash

ostt model select deepgram/nova-3

Per command:

bash

ostt transcribe meeting.mp3 -m deepgram/nova-3

Params

Persistent config:

toml

[deepgram.nova-3.params]
detect_language = ["en", "sv"]
diarize = true
smart_format = true
keyterm = ["OSTT", "Whisper", "Deepgram"]

Per invocation:

bash

ostt transcribe meeting.mp3 -m deepgram/nova-3 --param diarize=true --param smart_format=true --param detect_language=en,sv
ostt model params deepgram/nova-3 --format json

TOML lists must use TOML list syntax (["en", "sv"]). CLI --param list values use commas (--param detect_language=en,sv).

Params

Param	Type	Description
`detect_language`	boolean or string list	`true` detects the dominant language. A list restricts detection to given BCP-47 language codes, e.g. `["en", "sv"]`.
`language`	string	Primary language hint. Use when the language is known instead of detection.
`diarize`	boolean	Assign speaker numbers to words.
`diarize_model`	string	Select a batch diarization model: `latest`, `v1`, or `v2`. Do not combine with `diarize=true` unless Deepgram documents support for that combination.
`dictation`	boolean	Enable spoken formatting commands.
`filler_words`	boolean	Include filler words such as “uh” and “um”.
`measurements`	boolean	Convert spoken measurements to abbreviations.
`multichannel`	boolean	Transcribe each audio channel independently.
`numerals`	boolean	Convert spoken numbers to numeric form.
`paragraphs`	boolean	Add paragraph structure to the transcript.
`profanity_filter`	boolean	Mask or remove profanity.
`punctuate`	boolean	Add punctuation and capitalization.
`smart_format`	boolean	Format dates, times, currency, phone numbers, emails, URLs, and related text.
`utterances`	boolean	Segment speech into meaningful utterances.
`utt_split`	number	Pause duration in seconds before a new utterance.
`mip_opt_out`	boolean	Opt out of the Deepgram Model Improvement Program.
`keyterm`	string list	Nova-3 keyterm prompting. Saved `ostt keyword` terms are used as fallback only when `keyterm` is not set.
`keywords`	string list	Keyword boosting. Nova-2 uses this for saved `ostt keyword` fallback when `keywords` is not set.
`search`	string list	Search for terms in submitted audio.
`replace`	string list	Search and replace terms or phrases.
`redact`	string list	Redact sensitive information, such as `pci`, `pii`, or `numbers`.
`tag`	string list	Attach labels for reporting.
`extra`	string list	Attach arbitrary key-value metadata to the response.
`detect_entities`	boolean	Extract named entities.
`sentiment`	boolean	Detect sentiment.
`summarize`	boolean or string	Enable summarization. Deepgram also supports version strings such as `v2`.
`topics`	boolean	Detect topics.
`custom_topic`	string list	Topics to detect.
`custom_topic_mode`	string	Topic mode: `extended` or `strict`.
`intents`	boolean	Detect speaker intent.
`custom_intent`	string list	Intents to detect.
`custom_intent_mode`	string	Intent mode: `extended` or `strict`.
`encoding`	string	Expected submitted audio encoding, e.g. `linear16`, `flac`, `opus`. Usually unnecessary because OSTT submits encoded audio.
`version`	string	Model version, such as `latest`.

Deepgram ​

Models ​

Select A Model ​

Params ​

Params ​

Deepgram

Models

Select A Model

Params

Params