​
Overview

DeepgramSTTService provides real-time speech recognition using Deepgram’s WebSocket API with support for interim results, language detection, and voice activity detection (VAD).

API Reference

Complete API documentation and method details

Deepgram Docs

Official Deepgram documentation and features

Example Code

Working example with interruption handling

​
Installation

To use DeepgramSTTService, install the required dependencies: 
pip install "pipecat-ai[deepgram]"
You’ll also need to set up your Deepgram API key as an environment variable: DEEPGRAM_API_KEY.
Get your API key from the Deepgram Console.

​
Frames

​
Input

  • InputAudioRawFrame - Raw PCM audio data (16-bit, 16kHz, mono)
  • UserStartedSpeakingFrame - Triggers metrics collection
  • UserStoppedSpeakingFrame - Sends finalize command to flush session
  • STTUpdateSettingsFrame - Runtime transcription configuration updates
  • STTMuteFrame - Mute audio input for transcription

​
Output

  • InterimTranscriptionFrame - Real-time transcription updates
  • TranscriptionFrame - Final transcription results
  • ErrorFrame - Connection or processing errors

​
Models

Deepgram offers several models optimized for different use cases. Popular models include:
ModelBest ForFeatures
nova-3-generalGeneral purpose, meetingsLatest accuracy, punctuation
nova-2-generalGeneral purpose, meetingsLatest accuracy, punctuation
nova-2-phonecallPhone calls, low quality audioNoise robust, telephony optimized
See Deepgram’s model docs for detailed performance metrics.

​
Language Support

Deepgram STT supports the following languages and regional variants:
Language CodeDescriptionService Codes
Language.BGBulgarianbg
Language.CACatalanca
Language.ZHChinese (Mandarin, Simplified)zh, zh-CN, zh-Hans
Language.ZH_TWChinese (Mandarin, Traditional)zh-TW, zh-Hant
Language.ZH_HKChinese (Cantonese, Traditional)zh-HK
Language.CSCzechcs
Language.DADanishda, da-DK
Language.NLDutchnl
Language.NL_BEDutch (Flemish)nl-BE
Language.ENEnglishen
Language.EN_USEnglish (US)en-US
Language.EN_AUEnglish (Australia)en-AU
Language.EN_GBEnglish (UK)en-GB
Language.EN_NZEnglish (New Zealand)en-NZ
Language.EN_INEnglish (India)en-IN
Language.ETEstonianet
Language.FIFinnishfi
Language.FRFrenchfr
Language.FR_CAFrench (Canada)fr-CA
Language.DEGermande
Language.DE_CHGerman (Switzerland)de-CH
Language.ELGreekel
Language.HIHindihi
Language.HUHungarianhu
Language.IDIndonesianid
Language.ITItalianit
Language.JAJapaneseja
Language.KOKoreanko, ko-KR
Language.LVLatvianlv
Language.LTLithuanianlt
Language.MSMalayms
Language.NONorwegianno
Language.PLPolishpl
Language.PTPortuguesept
Language.PT_BRPortuguese (Brazil)pt-BR
Language.PT_PTPortuguese (Portugal)pt-PT
Language.RORomanianro
Language.RURussianru
Language.SKSlovaksk
Language.ESSpanishes, es-419
Language.SVSwedishsv, sv-SE
Language.THThaith, th-TH
Language.TRTurkishtr
Language.UKUkrainianuk
Language.VIVietnamesevi

​
Usage Example

​
Basic Configuration

Initialize the DeepgramSTTService and use it in a pipeline: 
from pipecat.services.deepgram.stt import DeepgramSTTService, LiveOptions
from pipecat.transcriptions.language import Language

# Configure service
stt = DeepgramSTTService(
    api_key="your-api-key",
    live_options=LiveOptions(
        model="nova-3-general",
        language=Language.EN,
        smart_format=True
    )
)

# Use in pipeline
pipeline = Pipeline([
    transport.input(),
    stt,
    context_aggregator.user(),
    llm,
    tts,
    transport.output(),
    context_aggregator.assistant()
])

​
Dynamic Configuration

Update settings dynamically by pushing an STTUpdateSettingsFrame: 
from pipecat.frames.frames import STTUpdateSettingsFrame

await task.queue_frame(STTUpdateSettingsFrame(
    language=Language.FR,
))

​
Metrics

The service provides:
  • Time to First Byte (TTFB) - Latency from audio input to first transcription
  • Processing Duration - Total time spent processing audio
Learn how to enable Metrics in your Pipeline.

​
Additional Notes

  • Connection Management: Automatically handles WebSocket connections and reconnections
  • VAD Integration: Supports Deepgram’s built-in VAD, though we recommend using local VAD services like Silero for better performance
  • Sample Rate: Can be configured per service, but we recommend setting it globally in PipelineParams for consistency across services