Deepgram | Nova-3 | Speech to Text Pro.

Deepgram Nova-3 Speech to Text Pro Overview

Deepgram Nova-3 Pro delivers advanced speech-to-text with built-in AI features like summarization, topic and entity detection, sentiment and intent analysis, plus smart formatting and redaction.

The model transcribes a media URL and exposes speaker, formatting and analysis controls in one request.

Deepgram Nova-3 Speech to Text Pro from Deepgram is available through each::labs in the nova-3 family. Required fields are media_url.

language_code: Language of the audio (BCP-47 code). diarize: Identify different speakers.

Use the Deepgram Nova-3 Speech to Text Pro API when you need this transformation or analysis as a repeatable step in an application or creative pipeline. The sections below distinguish the required material, adjustable controls and checks that matter before delivery.

Capabilities

• The model transcribes a media URL and exposes speaker, formatting and analysis controls in one request.
• language_code: Language of the audio (BCP-47 code).
• diarize: Identify different speakers.
• summarize: Generate a summary of the transcript.
• media_url: Audio file URL to transcribe.

Use Cases for Deepgram Nova-3 Speech to Text Pro

Deepgram Nova-3 Speech to Text Pro is useful when this specific operation fits between source preparation and delivery. These scenarios show how it can support different parts of a working pipeline.

Creators

Transcribe an interview with speaker labels. Use a representative source or a concise brief, then inspect the result before preparing a larger set of assets.

Design and marketing teams

Extract discussion topics from a call. Compare the result with the intended message, composition and destination. Keep reviewable versions of the source and output.

Developers

Prepare a readable transcript with selected redaction. Inspect the actual returned value before connecting it to another step, and keep the chosen request settings with the result.

Start with a clear definition of the desired result and use the exposed controls to reach it. Keep any additional generation, editing or application logic in separate steps so that the effect of this operation remains easy to review.

Tips and Tricks

Use clean source material where the request accepts audio or video. For a generative prompt, describe the desired sound, pacing and arrangement clearly. Change one exposed control at a time when comparing a result.

• diarize: Identify different speakers.
• summarize: Generate a summary of the transcript.

Set the named operation controls directly. This entry has no general creative-prompt interface; supplying an invented prompt field will not replace a required source or numeric setting.

Compare one deliberate adjustment at a time. Inspect the part of the result affected by that change, then check the rest of the asset or response for unwanted side effects before reusing the settings.

Technical Specifications

• Result type: object.
• Required fields: media_url.
• smart_format: boolean; default true.
• language_code: string; options auto, multi, tr, ur, en, nl, uk, es, ar, de, fr, it, ja, ko, pt, ru, zh, hi, bn, cs, da, fi, el, he, hu, id, ms, no, pl, ro, sk, sv, ta, te, th, vi; default "auto".
• diarize: boolean; default true.
• summarize: boolean; default true.

media_url: Audio file URL to transcribe.

Values must use the types and accepted options listed in the parameter reference. A displayed default does not remove a required-field requirement. Choose a supported output setting for this entry rather than borrowing a preset from a related model.

Things to Be Aware Of

Listen to the full result and inspect any returned structured data separately. Pronunciation, timing, unwanted sounds and transitions can require manual review even when a request completes successfully.

Optional settings are not necessarily interchangeable with those on another entry in the same family. Use the accepted names, types and values shown for this operation.

For media-based inputs, supply a directly retrievable file URL rather than a page containing the asset. Keep the source available while processing. Check the final content in the destination player, editor or application, not only the request status.

Key Considerations

Some options depend on others: paragraphs requires punctuation, utterance splitting requires utterances, and key terms are Nova-3-specific. Review transcription and extracted analysis independently.

Several pricing tiers apply. Check the full pricing reference for the chosen settings and supplied references rather than assuming one flat charge.

Prepare media_url before starting. Optional settings include language_code, diarize, summarize. Keep these settings aligned with the intended delivery, especially when a later step expects a particular type or format.

Run a representative request before scaling to a larger collection. Judge the result against the source and the brief, and account for all generated outputs or usage-sensitive options when estimating a batch.

Limitations

• language_code: allowed values: ["auto", "multi", "tr", "ur", "en", "nl", "uk", "es", "ar", "de", "fr", "it", "ja", "ko", "pt", "ru", "zh", "hi", "bn", "cs", "da", "fi", "el", "he", "hu", "id", "ms", "no", "pl", "ro", "sk", "sv", "ta", "te", "th", "vi"].
• redact: allowed values: ["pci", "pii", "phi", "numbers"].

Unsupported parameters do not establish additional capabilities. A generated result can differ from the intended brief, so review detail and consistency before publication.