Gnani

Speech-to-text services for Gnani Vachana — REST and WebSocket streaming for Indian languages

Overview

GnaniHttpSTTService and GnaniSTTService transcribe speech using Gnani Vachana, a production speech AI platform for 10+ Indian languages with real-time streaming and multilingual transcription.

Use GnaniHttpSTTService for REST-based file transcription of VAD-segmented utterances, or GnaniSTTService for low-latency WebSocket streaming with built-in VAD events. The WebSocket service emits TranscriptionFrame (final) and InterimTranscriptionFrame when the API sets is_final: false.

Source Repository

Source code, examples, and issues for the Gnani integration

PyPI Package

The pipecat-gnani package on PyPI

Gnani Website

Learn about Gnani’s speech AI platform

Gnani API Docs

STT REST and WebSocket API reference

Demo Video

Gnani × Pipecat integration — STT/TTS pipeline and interruption handling

gnani-vachana SDK

Core Gnani Vachana Python SDK (>= 0.7.9) installed as a dependency

Installation

Install the community package directly:

uv add pipecat-gnani

This also installs the gnani-vachana (>= 0.7.9) core SDK. The Python import package name remains gnani.

Prerequisites

Gnani Account Setup

Before using the Gnani STT services, you need:

1. Gnani API key: Get one from Gnani APIs.
2. Set credentials: Export GNANI_API_KEY as an environment variable.

Required Environment Variables

• GNANI_API_KEY: Your Gnani Vachana API key for authentication

Configuration

GnaniHttpSTTService (REST)

api_key · str · required

Gnani Vachana API key for authentication.

aiohttp_session · aiohttp.ClientSession · required

Shared aiohttp session for HTTP requests.

settings · GnaniHttpSTTService.Settings · default: "None"

Runtime-updatable STT settings. See Settings below.

GnaniHttpSTTService extends SegmentedSTTService and requires VAD in the pipeline so that speech segments are buffered and sent as whole utterances.

GnaniSTTService (WebSocket)

api_key · str · required

Gnani Vachana API key for authentication.

sample_rate · int · default: "None"

Audio sample rate in Hz. One of 8000, 16000, 44100, or 48000. When omitted, the rate is negotiated from StartFrame.

keepalive_timeout · float · default: "None"

Seconds of silence before sending keepalive audio. None disables keepalive.

keepalive_interval · float · default: "5.0"

Seconds between keepalive checks when keepalive is enabled.

settings · GnaniSTTService.Settings · default: "None"

Runtime-updatable STT settings. See Settings below.

Settings

Runtime-configurable settings passed via the settings constructor argument.

GnaniHttpSTTService.Settings

Parameter	Type	Default	Description
language	Language	Language.EN_IN	Recognition language (BCP-47, e.g. Language.HI_IN).
format	str	None	verbatim for raw output, transcribe for ITN-normalized output.
itn_native_numerals	bool	None	When format=transcribe, render digits in native script.

GnaniSTTService.Settings

Parameter	Type	Default	Description
language	Language	Language.EN_IN	Recognition language (BCP-47, e.g. Language.HI_IN).
sample_rate	int	None	Audio sample rate (8000, 16000, 44100, or 48000).
format	str	None	verbatim for raw output, transcribe for ITN-normalized output.
itn_native_numerals	bool	None	When format=transcribe, render digits in native script.

Supported STT languages include Assamese, Bengali, English (India), Gujarati, Hindi, Kannada, Malayalam, Marathi, Odia, Punjabi, Tamil, and Telugu. Changing language, format, itn_native_numerals, or sample_rate mid-session on the WebSocket service reconnects to re-run the init handshake. See the source repository for the authoritative, up-to-date list.

Streaming PCM specification

WebSocket STT sends raw PCM binary frames — no WAV or MP3 container mid-stream. Each frame is 1,024 bytes (512 samples × 2 bytes, 16-bit little-endian mono). Send frames at real-time cadence for best VAD accuracy.

Property	16 kHz	8 kHz
Encoding	PCM signed 16-bit little-endian	PCM signed 16-bit little-endian
Sample rate	16,000 Hz	8,000 Hz
Channels	1 (mono)	1 (mono)
Bytes per frame	1,024	1,024
Frame duration	32 ms	64 ms

GnaniSTTService also accepts 44,100 Hz and 48,000 Hz sample rates (x-sample-rate header). The server resamples internally — still send 1,024-byte frames at the appropriate cadence. See STT Realtime — PCM Specification.

Usage

Streaming (WebSocket)

import os

from pipecat_gnani import GnaniSTTService

stt = GnaniSTTService(
    api_key=os.getenv("GNANI_API_KEY"),
    sample_rate=16000,
    settings=GnaniSTTService.Settings(
        language=Language.HI_IN,
    ),
)

# ... add stt to your pipeline

REST (file-based)

import os

from pipecat_gnani import GnaniHttpSTTService

stt = GnaniHttpSTTService(
    api_key=os.getenv("GNANI_API_KEY"),
    aiohttp_session=session,
    settings=GnaniHttpSTTService.Settings(
        language=Language.HI_IN,
    ),
)

# ... add stt to your pipeline (VAD required)

Compatibility

Tested with the latest Pipecat main branch (v1.5.0+) using pipecat-gnani. Requires gnani-vachana >= 0.7.9. Check the source repository for the latest tested version and changelog.