text-to-speech

name	text-to-speech
description	Convert text to natural speech using Sarvam AI's Bulbul model. Use when the user needs to generate audio from text, create voiceovers, build voice interfaces, or synthesize Indian language speech. Supports 11 Indian languages with multiple voices, controllable pitch/pace/loudness, and real-time streaming. Returns base64-encoded audio.
license	Apache-2.0
metadata	{"author":"sarvam-ai","version":"1.0","model":"bulbul:v2"}

Text-to-Speech with Bulbul

Bulbul is Sarvam AI's text-to-speech model that generates natural-sounding speech in Indian languages with support for voice customization and streaming.

Installation

pip install sarvamai

Quick Start

from sarvamai import SarvamAI
from sarvamai.play import save

client = SarvamAI()

response = client.text_to_speech.convert(
    text="नमस्ते, आप कैसे हैं?",
    target_language_code="hi-IN",
    model="bulbul:v2",
    speaker="anushka"
)

# Response contains base64-encoded audio
save(response,
"output.wav")

Base64 Audio Response

The API returns audio as base64-encoded strings in the audios array:

{
  "request_id": "abc123",
  "audios": ["UklGRiQAAABXQVZFZm10IBAAAAABAAEA..."]
}

Decode Manually

import base64

response = client.text_to_speech.convert(
    text="Hello world",
    target_language_code="en-IN",
    model="bulbul:v2",
    speaker="anushka"
)

# Decode base64 to bytes
audio_bytes = base64.b64decode(response.audios[
    0
])

# Save to file
with open("output.wav",
"wb") as f:
    f.write(audio_bytes)

Supported Languages

Code	Language	Code	Language
hi-IN	Hindi	ta-IN	Tamil
bn-IN	Bengali	te-IN	Telugu
kn-IN	Kannada	ml-IN	Malayalam
mr-IN	Marathi	gu-IN	Gujarati
pa-IN	Punjabi	or-IN	Odia
en-IN	English (Indian)

Available Voices

Voice	Type	Best For
anushka	Female	General, warm tone
manisha	Female	Professional, clear
vidya	Female	Friendly, conversational
arjun	Male	Authoritative, news
amol	Male	Casual, storytelling
amartya	Male	Deep, formal

Voice Control

Customize pitch, pace, and loudness:

response = client.text_to_speech.convert(
    text="यह एक परीक्षण है।",
    target_language_code="hi-IN",
    model="bulbul:v2",
    speaker="anushka",
    pitch=0.2,          # -1.0 to 1.0 (higher = higher pitch)
    pace=1.2,           # 0.5 to 2.0 (higher = faster)
    loudness=1.5        # 0.5 to 2.0 (higher = louder)
)

Audio Formats

Set output format with output_audio_codec:

Format	Description
wav	Uncompressed (default)
mp3	MPEG Layer-3
aac	Advanced Audio Coding
opus	Optimized for speech
flac	Lossless
linear16	Raw PCM
mulaw	Telephony (8-bit)
alaw	Telephony (8-bit)

response = client.text_to_speech.convert(
    text="Hello",
    target_language_code="en-IN",
    model="bulbul:v2",
    speaker="anushka",
    output_audio_codec="mp3"
)

Sample Rates

Rate	Use Case
8000	Telephony
16000	Voice assistants
22050	Standard audio
24000	High quality (default)

response = client.text_to_speech.convert(
    text="Hello",
    target_language_code="en-IN",
    model="bulbul:v2",
    speaker="anushka",
    sample_rate=8000  # For phone systems
)

JavaScript

import { SarvamAI } from 'sarvamai'
import fs from 'fs'

const client = new SarvamAI()

const response = await client.textToSpeech.convert({
  text: 'नमस्ते',
  targetLanguageCode: 'hi-IN',
  model: 'bulbul:v2',
  speaker: 'anushka'
})

// Decode base64 and save
const audioBuffer = Buffer.from(response.audios[0], 'base64')
fs.writeFileSync('output.wav', audioBuffer)

cURL

curl -X POST "https://api.sarvam.ai/text-to-speech" \
  -H "api-subscription-key: $SARVAM_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "inputs": [
        "नमस्ते, कैसे हो?"
    ],
    "target_language_code": "hi-IN",
    "model": "bulbul:v2",
    "speaker": "anushka"
}'

Parameters

Parameter	Type	Required	Description
text / inputs	string/array	Yes	Text to synthesize
target_language_code	string	Yes	BCP-47 language code
model	string	Yes	bulbul:v2 or bulbul:v1
speaker	string	Yes	Voice name
pitch	float	No	-1.0 to 1.0
pace	float	No	0.5 to 2.0
loudness	float	No	0.5 to 2.0
output_audio_codec	string	No	Audio format
sample_rate	int	No	Output sample rate

Response

{
  "request_id": "20241115_abc123",
  "audios": ["UklGRiQAAABXQVZFZm10IBAAAAABAAEA..."]
}

See references/voices.md for voice samples and recommendations.