Sarvam AI STT

The SarvamSTTService provides specialized speech recognition for Indian languages using Sarvam AI's WebSocket API. It supports both standard STT (saarika models) and STT-Translate (saaras models) which auto-detect language and return translated text.

Installation

To use Sarvam AI, install the required dependencies:

pip install "piopiy-ai[sarvam]"

Prerequisites

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

  export SARVAM_API_KEY="your_api_key_here"

Configuration

SarvamSTTService Parameters

Parameter	Type	Default	Description
api_key	str	Required	Your Sarvam AI API key.
model	str	"saarika:v2.5"	Sarvam model to use (saarika or saaras).
input_audio_codec	str	"wav"	Audio format of the input.
params	InputParams	None	Configuration parameters (see below).

InputParams Parameters

Option	Type	Default	Description
language	Language	None	Target language (required for saarika).
prompt	str	None	Translation guide (only for saaras).
vad_signals	bool	True	Whether to enable VAD signals.
high_vad_sensitivity	bool	False	Enable high VAD sensitivity.

Usage

Basic STT (Saarika)

import os
from piopiy.services.sarvam.stt import SarvamSTTService
from piopiy.transcriptions.language import Language

stt = SarvamSTTService(
    api_key=os.getenv("SARVAM_API_KEY"),
    model="saarika:v2.5",
    params=SarvamSTTService.InputParams(
        language=Language.HI_IN # Hindi
    )
)

STT-Translate (Saaras)

import os
from piopiy.services.sarvam.stt import SarvamSTTService

stt = SarvamSTTService(
    api_key=os.getenv("SARVAM_API_KEY"),
    model="saaras:v1",
    params=SarvamSTTService.InputParams(
        prompt="Translate this customer query into clear English."
    )
)

Notes

• Indian Language Focus: Sarvam is purpose-built to excel at regional Indian accents and dialects.
• Model Types: saarika models require a fixed language, while saaras models perform automatic language detection and translation.