Gnani

Text-to-speech services for Gnani Vachana — REST, SSE, and WebSocket streaming for Indian languages

Overview

GnaniHttpTTSService, GnaniSSETTSService, and GnaniTTSService synthesize speech using Gnani Vachana, a production speech AI platform for 10+ Indian languages with real-time streaming and low-latency synthesis.

Use the REST service for simple request/response synthesis, the SSE service for streaming with lower latency than REST, or the WebSocket GnaniTTSService for lowest latency with interruption support via InterruptibleTTSService.

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

TTS REST, SSE, 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 TTS 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

All three services share the same audio-format settings via their Settings dataclasses. The key difference is transport: REST for single-request synthesis, SSE for chunked streaming, and WebSocket for interruptible real-time synthesis.

Shared constructor parameters

api_key · str · required

Gnani Vachana API key for authentication.

voice_id · str · default: "Pranav"

Voice name for synthesis. For timbre-v2.0: Pranav, Kaveri, Shubhra, Deepak. For timbre-v2.5: 42 voices across 11 language groups — see Available voices below.

model · str · default: "timbre-v2.0"

TTS model identifier. One of timbre-v2.0 or timbre-v2.5.

sample_rate · int · default: "None"

Output sample rate in Hz. One of 8000, 16000, 22050, 24000, 44100, or 48000. When omitted, negotiated from StartFrame.

settings · Settings · default: "None"

Runtime-updatable TTS settings. See Settings below.

GnaniHttpTTSService and GnaniSSETTSService (HTTP)

aiohttp_session · aiohttp.ClientSession · required

Shared aiohttp session for HTTP requests.

Settings

Runtime-configurable settings passed via the settings constructor argument using GnaniHttpTTSService.Settings(...), GnaniSSETTSService.Settings(...), or GnaniTTSService.Settings(...).

Parameter	Type	Default	Description
voice	str	"Pranav"	Voice identifier. See Available voices.
model	str	"timbre-v2.0"	TTS model (timbre-v2.0 or timbre-v2.5).
language	Language	None	Synthesis language (timbre-v2.5 only). BCP-47 code, e.g. Language.HI_IN.
encoding	str	"linear_pcm"	Audio encoding (linear_pcm, oggopus, pcm_mulaw, pcm_alaw).
container	str	"wav"	Audio container (raw, mp3, wav, ogg, mulaw, alaw).
sample_width	int	2	Sample width in bytes (1–4).
bitrate	str	None	MP3 bitrate (32k, 64k, 96k, 128k, 192k). Only used when container=mp3.

For timbre-v2.0, voice selection determines synthesis language — no language parameter is sent. For timbre-v2.5, set language explicitly to control the synthesis language (e.g. Language.HI_IN for Hindi).

When using encoding="oggopus", set container="raw" (the API rejects oggopus with container="ogg"). For telephony A-law output, use container="alaw" with sample_rate=8000.

Voice, model, and audio-format settings take effect on the next synthesis request. See the source repository and Gnani TTS docs for the authoritative, up-to-date voice and language lists.

Available voices

timbre-v2.0 (4 voices)

Pranav, Kaveri, Shubhra, Deepak

timbre-v2.5 (42 voices)

Language	Voices
Hindi	Nalini, Bhavna, Yashvi, Urmila, Jwala, Chitra, Ambuja, Deepak, Roopesh, Vikrant, Hemraj, Jalaj, Omkar
English	Kaveri, Trupti, Devika, Pranav, Shlok, Girish
Tamil	Asmita, Trisha, Brinda, Vedika, Noopur
Telugu	Suhana, Lehara, Lavanya, Yukti, Varuni
Kannada	Saanvi, Kavin
Malayalam	Reshma, Riyaan
Marathi	Zahira, Ishaan
Bengali	Kirra, Dhruva
Gujarati	Falak, Veera
Punjabi	Mehuli, Zayan
Hinglish	Poorvi

timbre-v2.5 requires a language parameter (BCP-47 code, e.g. hi-IN) when using multilingual voices. Some voices (like Pranav, Kaveri, Deepak) appear in both models — select the model to pick the voice variant.

Usage

Streaming with interruption (WebSocket)

import os

from pipecat_gnani import GnaniTTSService

tts = GnaniTTSService(
    api_key=os.getenv("GNANI_API_KEY"),
    sample_rate=16000,
    settings=GnaniTTSService.Settings(
        voice="Pranav",
    ),
)

# ... add tts to your pipeline

Using timbre-v2.5 with language

import os

from pipecat_gnani import GnaniTTSService

tts = GnaniTTSService(
    api_key=os.getenv("GNANI_API_KEY"),
    sample_rate=16000,
    settings=GnaniTTSService.Settings(
        model="timbre-v2.5",
        voice="Nalini",
        language=Language.HI_IN,
    ),
)

# ... add tts to your pipeline

SSE streaming

import os

from pipecat_gnani import GnaniSSETTSService

tts = GnaniSSETTSService(
    api_key=os.getenv("GNANI_API_KEY"),
    aiohttp_session=session,
    settings=GnaniSSETTSService.Settings(
        voice="Pranav",
    ),
)

# ... add tts to your pipeline

REST (non-streaming)

import os

from pipecat_gnani import GnaniHttpTTSService

tts = GnaniHttpTTSService(
    api_key=os.getenv("GNANI_API_KEY"),
    aiohttp_session=session,
    settings=GnaniHttpTTSService.Settings(
        voice="Pranav",
    ),
)

# ... add tts to your pipeline

Minimal pipeline example

import os

from pipecat.pipeline.pipeline import Pipeline
from pipecat.transcriptions.language import Language
from pipecat_gnani import GnaniSTTService, GnaniTTSService

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

tts = GnaniTTSService(
    api_key=os.getenv("GNANI_API_KEY"),
    settings=GnaniTTSService.Settings(voice="Pranav"),
)

pipeline = Pipeline(
    [
        transport.input(),
        stt,
        # ... LLM and aggregators
        tts,
        transport.output(),
    ]
)

Compatibility

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