is it possible to do context best formatting with streaming stt?

[waxen egret]

is it possible to do context best formatting with streaming stt? like I will do streaming stt and based on context the output will be format accordingly?

[modern sequoia]

Hey there! This channel allows you to ask Kapa.ai for assistance. But it looks like you haven't connected your Discord profile to your Deepgram account yet. You can do so at https://community.deepgram.com - by doing so, you'll be able to ask questions in #1115960287183511643 if you need human assistance!

[sage crane]

Hi @waxen egret. I'm kapa.ai, your friendly support bot. I will try to answer your question by searching through the knowledge base. :wave:

---

Yes, you can get context-based (“best effort”) formatting with Deepgram’s streaming STT, but there are some details and limits.

1. Smart Formatting in Streaming

Deepgram’s smart_format works for both pre-recorded and streaming STT. When enabled, it applies context-aware formatting such as punctuation, paragraphs, and (especially for English) things like dates, times, currency, phone numbers, emails, and URLs. [Smart format; Streaming features]

For streaming, you just add:

smart_format=true

For example, with the JS SDK:

const connection = deepgram.listen.live({
  model: "nova-3",
  language: "en-US",
  smart_format: true,
});

[Live streaming]

2. Context-Aware Finalization Behavior

With streaming + smart formatting, Deepgram waits briefly to gather enough context around entities (like numbers or dates) before finalizing and formatting them:

• It waits until the speaker continues to non-entity speech, or
• Finalizes after about 3 seconds of silence
  and then formats based on the available context. [Smart format]

You can override this behavior with:

• no_delay=true → forces immediate finalization but often skips formatting. [No delay]

3. Voice Agent STT Context Formatting

If you’re using the Voice Agent API with Deepgram as the listen provider, there’s also:

agent.listen.provider.smart_format: true

This is specifically to improve transcript readability for UI or downstream processing in a live agent setting. [Voice agent smart_format]

4. Self-Hosted / NER-Based Formatting (Advanced)

For self-hosted Nova-3 or Nova-2, you can enable NER-based formatting (Named Entity Recognition), which uses an entity detector plus non-formatted models to apply human-like formatting (e.g., turning “six zero eight…” into [phone]). [Self-hosted NER]

That is also context-aware, but it requires:

• Specific self-hosted models (entity-detector + non-formatted STT)
• Feature flags like streaming_ner = true in your config

This is not exposed as a separate cloud parameter beyond smart_format; it’s part of how newer models handle formatting.

---

So in summary:

• Yes, Deepgram streaming STT supports context-based formatting via smart_format=true.
• Formatting decisions are context-sensitive and tied to streaming finalization timing.
• You can trade off latency vs. formatting quality using the default behavior vs. no_delay=true.

---

Note: you can ask me a follow up question by @ mentioning me again :speech_balloon: