Deepgram STT

The Deepgram STT provider enables your agent to use Deepgram's advanced speech-to-text models for high-accuracy, real-time audio transcription.

Installation

Install the Deepgram-enabled VideoSDK Agents package:

pip install "videosdk-plugins-deepgram"

Importing

from videosdk.plugins.deepgram import DeepgramSTT

Example Usage

from videosdk.plugins.deepgram import DeepgramSTT
from videosdk.agents import CascadingPipeline

# Initialize the Deepgram STT model
stt = DeepgramSTT(
    # When DEEPGRAM_API_KEY is set in .env - DON'T pass api_key parameter
    api_key="your-deepgram-api-key",
    model="nova-2",
    language="en-US",
    interim_results=True,
    punctuate=True,
    smart_format=True
)

#  Add stt to cascading pipeline
pipeline = CascadingPipeline(stt=stt)

note

When using .env file for credentials, don't pass them as arguments to model instances or context objects. The SDK automatically reads environment variables, so omit api_key, videosdk_auth, and other credential parameters from your code.

Configuration Options

• api_key: Your Deepgram API key (can also be set via environment variable)
• model: The Deepgram model to use (e.g., "nova-2", "nova-3", "whisper-large")
• language: (str) Language code for transcription (default: "en-US")
• interim_results: (bool) Enable real-time partial transcription results (default: True)
• punctuate: (bool) Add punctuation to transcription (default: True)
• smart_format: (bool) Apply intelligent formatting to output (default: True)
• sample_rate: (int) Audio sample rate in Hz (default: 48000)
• endpointing: (int) Silence detection threshold in milliseconds (default: 50)
• filler_words: (bool) Include filler words like "uh", "um" in transcription (default: True)
• base_url: (str) WebSocket endpoint URL (default: "wss://api.deepgram.com/v1/listen")