Gladia

​

Overview

GladiaSTTService is a speech-to-text (STT) service that integrates with Gladia’s API to provide real-time transcription capabilities. It processes audio input and produces transcription frames in real-time.

​

Installation

To use GladiaSTTService, you need to install the Gladia dependencies:

pip install pipecat-ai[gladia]

You’ll also need to set up your Gladia API key as an environment variable: GLADIA_API_KEY

​

Configuration

​

Service Parameters

​

Audio Processing Parameters

​

Input

The service processes raw audio data with the following requirements:

• PCM audio format
• 16-bit depth
• 16kHz sample rate (default)
• Single channel (mono)

​

Output

The service produces two types of frames during transcription:

​

TranscriptionFrame

Generated for final transcriptions, containing:

​

InterimTranscriptionFrame

Generated during ongoing speech, containing the same fields as TranscriptionFrame but with preliminary results.

​

ErrorFrame

Generated when transcription errors occur, containing error details.

​

Methods

See the STT base class methods for additional functionality.

​

Language Setting

await service.set_language(Language.FR)

​

Language Support

Gladia STT supports the following languages:

​

Usage Example

from pipecat.services.gladia import GladiaSTTService
from pipecat.transcriptions.language import Language

# Configure the service
stt_service = GladiaSTTService(
    api_key="your-api-key",
    confidence=0.7,
    params=GladiaSTTService.InputParams(
        language=Language.EN,
        audio_enhancer=True,
        sample_rate=16000
    )
)

# Use in a pipeline
pipeline = Pipeline([
    transport.input(),    # Produces InputAudioRawFrame
    stt_service,          # Processes audio → produces transcription frames
    llm_processor,        # Consumes TranscriptionFrame
])

Note: Gladia uses simplified language codes without regional variants.

​

Frame Flow

​

Metrics Support

The service collects processing metrics:

• Time to First Byte (TTFB)
• Processing duration
• Connection status

​

Notes

• Audio input must be in PCM format
• Transcription frames are only generated when confidence threshold is met
• Language detection is optional
• Service automatically handles websocket connections and cleanup
• Real-time processing occurs in parallel for natural conversation flow