Switching the Speech-to-Text (STT)¶

GhostBrain uses Deepgram (DeepgramSTTService) by default because it offers ultra-low latency streaming and excellent accuracy. However, you can easily swap this out for another provider supported by Pipecat (like Gladia or Azure).

Step 1: Update Dependencies¶

First, check if Pipecat requires an optional dependency for your new provider. For example, to switch to Gladia:

# Optional: Add gladia extra dependency
hatch run pip install "pipecat-ai[gladia]"

Note: Update your pyproject.toml dependencies to make this permanent!

Step 2: Add the API Key¶

Add the new API key to your environment and Settings model.

1. In .env:

   GHOST_BRAIN_GLADIA_API_KEY=your_key_here
   

2. In src/ghost_brain/config.py:

   class Settings(BaseSettings):
       # ... existing keys ...
       gladia_api_key: str = Field(default="")
   

Step 3: Modify the STT Factory¶

Open src/ghost_brain/services/stt.py. Change the import and the returned class:

# Remove Deepgram import
# from pipecat.services.deepgram import DeepgramSTTService

# Add new import
from pipecat.services.gladia import GladiaSTTService
from ghost_brain.config import Settings

def create_stt(settings: Settings, sample_rate: int = 16000):
    """
    Create STT service (swapped to Gladia).
    """
    return GladiaSTTService(
        api_key=settings.gladia_api_key,
        # Gladia-specific configurations
        audio_config={
            "sample_rate": sample_rate,
            "encoding": "linear16"
        }
    )

Because Pipecat standardizes the interfaces for all its components, you don't need to change pipeline.py or any of the WebSocket logic! The pipeline will treat the new STT service exactly like the old one.