Sarvam

Updated

Integrate Sarvam TTS with the Conversational AI Engine.

Sarvam provides fast, reliable text-to-speech (TTS) with support for Indian languages and customizable voices.

Sample configuration

The following examples show how to configure Sarvam TTS when starting a conversational AI agent.

from agora_agent import Agent
from agora_agent.agentkit.vendors import SarvamTTS

# client is your configured Agora client
agent = (
    Agent(client)
    .with_stt(...)  # configure your STT vendor
    .with_llm(...)  # configure your LLM vendor
    .with_tts(SarvamTTS(
        key='your-sarvam-key',
        speaker='anushka',
        target_language_code='en-IN',
        model='bulbul:v3',
        speech_sample_rate=24000,
        enable_preprocessing=True,
    ))
)
import { Agent, SarvamTTS } from 'agora-agents';

// client is your configured Agora client
const agent = new Agent({ client })
  .withStt(/* configure your STT vendor */)
  .withLlm(/* configure your LLM vendor */)
  .withTts(new SarvamTTS({
    key: 'your-sarvam-key',
    speaker: 'anushka',
    targetLanguageCode: 'en-IN',
    model: 'bulbul:v3',
    speechSampleRate: 24000,
    enablePreprocessing: true,
  }));
import (
    "github.com/AgoraIO/agora-agents-go/v2/agentkit"
    "github.com/AgoraIO/agora-agents-go/v2/agentkit/vendors"
)

speechSampleRate := 24000
enablePreprocessing := true

// client is your configured Agora client
agent := agentkit.NewAgent(client).WithStt(/* configure your STT vendor */).
  WithLlm(/* configure your LLM vendor */).
  WithTts(
    vendors.NewSarvamTTS(vendors.SarvamTTSOptions{
        Key:                 "your-sarvam-key",
        Speaker:             "anushka",
        TargetLanguageCode:  "en-IN",
        Model:               "bulbul:v3",
        SpeechSampleRate:    &speechSampleRate,
        EnablePreprocessing: &enablePreprocessing,
    }),
)

Use the following tts configuration in your request:

"tts": {
  "vendor": "sarvam",
  "params": {
    "api_subscription_key": "<SARVAM_API_KEY>",
    "speaker": "anushka",
    "target_language_code": "en-IN",
    "model": "bulbul:v3",
    "speech_sample_rate": 24000,
    "enable_preprocessing": true
  }
}

Key parameters

params required
ParameterDescription
api_subscription_keystring
required

The API key used for authentication. Get your API key from the Sarvam dashboard.

speakerstring
required

The voice ID to use for speech generation.

  • Female voices: anushka, manisha, vidya, arya.
  • Male voices: abhilash, karun, hitesh.
target_language_codestring
required

The language code for speech synthesis. Supported values:

  • en-IN English (India)
  • hi-IN: Hindi
  • bn-IN: Bengali
  • ta-IN: Tamil
  • te-IN: Telugu
  • kn-IN: Kannada
  • ml-IN: Malayalam
  • mr-IN: Marathi
  • gu-IN: Gujarati
  • pa-IN: Punjabi
  • or-IN: Odia
pitchnumber
optional
Possible values
  • [-0.75,0.75]

The pitch adjustment for the voice. Positive values make the voice sharper; negative values make it deeper. Applies to the bulbul:v2 model only.

pacenumber
optional
Possible values
  • [0.3,3.0]

The speed of speech. 1.0 is the normal speed. Values less than 1.0 slow down the speech; values greater than 1.0 speed it up.

loudnessnumber
optional
Possible values
  • [0.1,3.0]

The volume level of the speech. Higher values increase the loudness. Applies to the bulbul:v2 model only.

modelstring
optional
Default value
bulbul:v3
Possible values
  • bulbul:v2
  • bulbul:v3

The Sarvam TTS model to use.

speech_sample_ratenumber
optional
Default value
24000
Possible values
  • 8000
  • 16000
  • 22050
  • 24000
  • 32000
  • 44100
  • 48000

The audio sample rate in Hz. Replaces sample_rate, which is deprecated.

enable_preprocessingboolean
optional

Whether to normalize English words and numeric entities, such as dates and currency, before speech synthesis.

The table extends beyond the current width. Scroll horizontally to view more columns.

Caution

The parameters listed on this page are validated for use with Conversational AI Engine. Required parameters must be provided as documented. Any additional parameters are passed through directly to the underlying vendor without validation. For a full list of supported options, refer to the Sarvam documentation.

This integration is fully supported for use with Conversational AI Engine. While it has completed functional validation, it is newer to the platform, and additional provider-specific edge cases may be identified as usage scales across a broader range of applications and workloads.