Soniox

The SonioxSTTService offers real-time speech-to-text with a focus on high precision and structured output. It supports speaker diarization and custom context for domain-specific vocabulary.

Installation

To use Soniox, install the required dependencies:

pip install "piopiy-ai[soniox]"

Prerequisites

• A Soniox account and API key (Get yours here).
• Set your API key in your environment:

  export SONIOX_API_KEY="your_api_key_here"

Configuration

SonioxSTTService Parameters

Parameter	Type	Default	Description
api_key	str	Required	Your Soniox API key.
sample_rate	int	None	Audio sample rate in Hz.
params	SonioxInputParams	None	Model and diarization settings.

SonioxInputParams

Parameter	Type	Default	Description
model	str	"stt-rt-preview"	Model ID for transcription.
language_hints	List[Language]	None	Hints to guide language detection.
context	SonioxContextObject	None	Custom vocabulary or phrases.
enable_speaker_diarization	bool	False	Enable speaker identification.

Usage

Basic Setup

import os
from piopiy.services.soniox.stt import SonioxSTTService, SonioxInputParams

stt = SonioxSTTService(
    api_key=os.getenv("SONIOX_API_KEY"),
    params=SonioxInputParams(
        model="stt-rt-preview",
        enable_speaker_diarization=True
    )
)

Notes

• Contextual Accuracy: Use the context parameter to provide specific terms (e.g., product names, technical jargon) to improve transcription accuracy in specialized domains.
• Diarization: Soniox provides per-token speaker identification, allowing for precise attribution of speech in multi-party conversations.