# Nova-3 Multilingual Speech to Text

Transcribe audio and video to text with Nova-3 Multilingual Speech to Text from Deepgram through one Spiral API key. Transcribes recorded audio where speakers switch between languages, with Deepgram Nova-3 multilingual (language=multi). Returns a JSON transcript with the language of each word; language_hint biases detection. It runs in the background and returns structured JSON. Pay as you go, with no subscription.

- Provider: Deepgram
- Category: Audio
- Runs: In the background
- Returns: json
- Released: 2025-03-31
- Tool id: `deepgram` / `nova-3-multilingual`
- Page: https://spiral.ai/tools/deepgram/nova-3-multilingual

## Pricing

$0.00676 per minute of audio, per channel. Redaction: $0.0026 per minute. Key terms: $0.00169 per minute. Entity detection: $0.00221 per minute.

## Limits

- Longest queue wait: 1 hour
- Run deadline: 30 minutes
- Longest run, end to end: 930 minutes
- Largest inline input: 1 MB
- Largest input file: 2 GB
- Largest output: 5 GB

## Input

| Field | Kind | Allowed | Default | Description |
| --- | --- | --- | --- | --- |
| `url` (required) | URL or file |  | — | Audio file URL to transcribe |
| `extra` | JSON |  | — | Arbitrary key-value pairs that are attached to the API response for usage in downstream processing |
| `redact` | JSON |  | — | Redaction removes sensitive information from your transcripts |
| `search` | JSON |  | — | Search for terms or phrases in submitted audio |
| `diarize` | true or false |  | — | Deprecated: use `diarize_model` instead. Recognize speaker changes. Each word in the transcript will be assigned a speaker number starting at 0. Cannot be sent together with `diarize_model`. |
| `keyterm` | list |  | — | Key term prompting improves recognition of specialized terminology and brands. |
| `replace` | JSON |  | — | Search for terms or phrases in submitted audio and replaces them |
| `encoding` | one of | linear16, flac, mulaw, amr-nb, amr-wb, opus, speex, g729 | — | Specify the expected encoding of your submitted audio |
| `keywords` | JSON |  | — | Keywords can boost or suppress specialized terminology and brands. `keywords` is not supported with Nova-3 models; use `keyterm` instead. |
| `numerals` | true or false |  | false | Numerals converts numbers from written format to numerical format |
| `dictation` | true or false |  | false | Dictation mode for controlling formatting with dictated speech |
| `punctuate` | true or false |  | — | Add punctuation and capitalization to the transcript Left unset, smart_format adds punctuation; false turns it off even with smart_format. |
| `utt_split` | number |  | 0.8 | Seconds to wait before detecting a pause between words in submitted audio |
| `paragraphs` | true or false |  | false | Splits audio into paragraphs to improve transcript readability |
| `utterances` | true or false |  | false | Segments speech into meaningful semantic units |
| `filler_words` | true or false |  | false | Filler Words can help transcribe interruptions in your audio, like "uh" and "um" |
| `measurements` | true or false |  | false | Spoken measurements will be converted to their corresponding abbreviations |
| `multichannel` | true or false |  | false | Transcribe each audio channel independently |
| `smart_format` | true or false |  | false | Apply formatting to transcript output. When set to true, additional formatting will be applied to transcripts to improve readability |
| `diarize_model` | one of | latest, v1, v2 | — | Select and enable a specific diarization model version. |
| `language_hint` | JSON |  | — | Biases language detection toward one or more languages on Nova-3 multilingual (`language` = `multi`); ignored with any other `language`. Pass a list for several codes. A code that is not in the Nova-3 multilingual language list fails the request. |
| `detect_entities` | true or false |  | false | Identifies and extracts key entities from content in submitted audio |
| `detect_language` | JSON |  | — | Identifies the dominant language spoken in submitted audio |
| `profanity_filter` | true or false |  | false | Profanity Filter looks for recognized profanity and converts it to the nearest recognized non-profane word or removes it from the transcript completely |

## Example

```bash
curl https://api.spiral.ai/v1/runs \
  -H "Authorization: Bearer $SPIRAL_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: $(uuidgen)" \
  -d '{
    "provider": "deepgram",
    "model": "nova-3-multilingual",
    "input": {
      "url": "<url>"
    }
  }'
```

Try it in the Playground: https://spiral.ai/start?provider=deepgram&model=nova-3-multilingual

## Common questions

### Nova-3 Multilingual Speech to Text

**How much does Nova-3 Multilingual Speech to Text cost?**

$0.00676 per minute of audio, per channel. Redaction: $0.0026 per minute. Key terms: $0.00169 per minute. Entity detection: $0.00221 per minute. You pay as you go from one Spiral balance, with no subscription, and a failed run is not charged.

**How do I use Nova-3 Multilingual Speech to Text through an API?**

Send `POST /v1/runs` with `"provider": "deepgram"`, `"model": "nova-3-multilingual"` and the input fields on this page, using one Spiral API key. It runs in the background: poll the run, or name a webhook to be told when it finishes.

**Do I need an account with Deepgram to use Nova-3 Multilingual Speech to Text?**

No. Spiral runs it on its own Deepgram account and charges your Spiral balance.

**Can my AI agent use Nova-3 Multilingual Speech to Text?**

Yes. Agents find and run it through Spiral's MCP server, CLI or agent skill, with the same API key.

