Audio Intelligence

Deepgram Audio Intelligence

Audio intelligence is not a separate endpoint. It is five query parameters on /v1/listen, the same endpoint that returns the transcript. You get the transcript and the analysis in one response, from one API call. This skill gets a verified request working and states the limits precisely.

Decide first

• Your input is audio (a file, a URL, a call recording) and you want analysis: stay here, use POST https://api.deepgram.com/v1/listen.
• Your input is already text (a transcript, an email, a chat log, a support ticket): the parameters below do not apply. Use the Read API, POST /v1/read. Open the text-intelligence skill.
• You only want the transcript: drop these parameters and open the speech-to-text skill.
• You are streaming live audio: only detect_entities is available. See the matrix.

Feature matrix

detect_entities is the odd one out twice over: it is the only feature that works on the live socket, and it is the only one that does not exist on the Read API. Streaming entity detection runs on Nova, Nova-2, Nova-3, and Enhanced; it is not available on Base models or on Flux. [2]

Verified request

curl -s -X POST 'https://api.deepgram.com/v1/listen?model=nova-3&smart_format=true&summarize=v2&sentiment=true&topics=true&intents=true&detect_entities=true' \
  -H "Authorization: Token $DEEPGRAM_API_KEY" \
  -H 'Content-Type: application/json' \
  -d '{"url":"https://dpgr.am/spacewalk.wav"}'

Returns 200. The analysis is scattered across the response, not collected in one place:

Note the plural: the parameter is sentiment, the result key is sentiments. metadata gains a summary_info, sentiment_info, topics_info, and intents_info block per enabled feature, each with model_uuid, input_tokens, and output_tokens. Entity labels come back uppercased (NAME, ORGANIZATION, LOCATION_CITY, MONEY, DATE_INTERVAL); Deepgram documents over 50 types. [6]

On the live socket, detect_entities=true adds a top-level entities array to each Results message, next to channel — not inside channel.alternatives[0]. Same field shape as above.

Narrowing topics and intents

custom_topic and custom_intent (repeatable) add your own labels; custom_topic_mode and custom_intent_mode take extended (default, your labels plus the model’s) or strict (your labels only). strict returns "segments": [] whenever nothing matches your list, which looks like a broken request but is not. Start with extended.

Common mistakes

1. Expecting these to work on streaming. Each failure mode is different, which makes this confusing. summarize fails the WebSocket handshake with 400 "Summarization is not available for streaming.". topics and intents fail it with 403 {"err_code":"UNAUTHORIZED_FEATURES_REQUESTED","err_msg":"Project does not have access to the requested feature/s [\"topics\"]."} — which reads as a permissions problem but is not one: the same key’s prerecorded topics requests succeed, and the docs matrix lists streaming as unsupported for both. [2] Do not go asking for an entitlement. sentiment is worse still: the handshake succeeds and no sentiment is ever returned. Only detect_entities works.
2. Expecting a non-English request to fail loudly. It does not. With language=es on real Spanish audio, sentiment, topics, intents, and detect_entities return HTTP 200 with the transcript, the analysis keys silently absent, and the reason only in metadata.warnings: [{"parameter":"sentiment","type":"unsupported_language","message":"Sentiment is only supported for English."}], plus "Topics are only supported for English.", "Intents are only supported for English.", and "Entity detection is only supported for English.". Read metadata.warnings before you conclude the model found nothing.
3. Assuming summarize behaves the same way. It is the exception: non-English is a hard 400, {"err_code":"Bad Request","err_msg":"Summarization v2 not supported for non-English languages"}. language=multi gets the same 400.
4. language=multi as a workaround. It is not one. multi returns the analysis when the detected speech is English and drops it with the same metadata.warnings when it is not, so the same request succeeds or silently degrades depending on what the caller said.
5. summarize=v1. Returns 400 "To use the summarize feature, please use 'summarize=true' or 'summarize=v2'. The 'summarize=v1' parameter is deprecated." Use v2; true is accepted and returns the same summary.short shape.
6. Looking for results.summary.text. That is the Read API’s shape. On /v1/listen the summary is at .

Pricing

Enabling these features changes what a request costs. Rates and the billing model change, so read https://deepgram.com/pricing (opens in a new tab) rather than any figure quoted in a skill.

Use a different skill when

• Your input is text rather than audio: text-intelligence skill (/v1/read).
• You want every parameter and the full response schema: api skill, references/listen.md.
• You only need transcription, diarization, redaction, or captions: speech-to-text skill.
• You want a runnable demo app: starters skill. Note there is no audio-intelligence starter; the text-intelligence feature (13 languages) is the Read API app.
• You want a snippet under 50 lines: recipes skill, “Audio Intelligence v1“ — summarize, sentiment, topics, intents, entities, in Python, JavaScript, Go, .NET, Java, Rust, and the CLI. [7]
• You want language-idiomatic SDK code: install deepgram-{js,python,java,go,rust,dotnet}-audio-intelligence from the matching SDK repository (npx skills add deepgram/deepgram-python-sdk, and so on).
• You want to find a docs page: docs skill. You want the docs in your editor: setup-mcp skill.

Sources

1. https://developers.deepgram.com/docs/audio-intelligence (opens in a new tab) (getting started)
2. https://developers.deepgram.com/docs/stt-intelligence-feature-overview (opens in a new tab) (the prerecorded/streaming/language matrix, and the streaming entity-detection model footnote)
3. https://developers.deepgram.com/docs/summarization (opens in a new tab), https://developers.deepgram.com/docs/sentiment-analysis (opens in a new tab), https://developers.deepgram.com/docs/topic-detection,