Mira TTS
₹1.80 per 1,000 characters.⁴
On pay as you go and on Business, paid from one rupee wallet. No plan to pick first.
Google Cloud Text-to-Speech alternative
They slip into English and back the way people talk on calls.
| Google Cloud Text-to-Speech | Voice Infra |
|---|---|
| Published price | |
| ₹2,849per 10 lakh characters, Chirp 3: HD (US$30) | ₹1,800per 10 lakh characters |
| Voices | |
| 30 Chirp 3: HD for Hindi, and more | Four, built for Hinglish calls |
| Phone-line audio | |
| μ-law and A-law | 8 kHz μ-law and A-law |
| Streaming | |
| Two-way, in Preview, Chirp 3: HD only | HTTP, and a WebSocket for a whole call |
Side by side
| Option | Bar | ₹ per 10 lakh characters |
|---|---|---|
| ₹2,849$30 | ||
| ₹1,519$16 | ||
| ₹380$4 | ||
| ₹1,800 |
Dollar prices converted at ₹94.96, the ECB reference rate on 16 October 2026. The rate.
After each month's free characters. 10 lakh is 1 million. Gemini-TTS is priced by tokens, so it isn't drawn.
From cloud.google.com/text-to-speech/pricing, checked . Ours from our price list.
| Detail | Google Cloud Text-to-Speech | Voice Infra |
|---|---|---|
| Published price | Chirp 3: HD voices US$30 per 1M characters Neural2 US$16; WaveNet and Standard US$4; Gemini-TTS priced per token cloud.google.com | ₹1.80 per 1,000 characters, on pay as you go and on Business (₹50,000/month commitment), before GST |
Google counts every character, including spaces and most SSML tags. Chirp 3: HD is ₹2,849 per million characters after a free first million each month. Neural2 is ₹1,519. WaveNet and Standard are ₹380.
Mira TTS is ₹1.80 per 1,000 characters, so a million cost ₹1,800. Every response tells you how many characters were billed, and a request that fails is free.
So Google’s older voices cost less than ours, and Chirp 3: HD costs more. Pick by how the voice sounds on your own lines, then by price.
| Detail | Google Cloud Text-to-Speech | Voice Infra |
|---|---|---|
| Voices | hi-IN: 30 Chirp 3: HD voices, 4 Neural2, 6 WaveNet and 6 Standard 30 Chirp 3: HD voices each for Gujarati, Tamil, Telugu, Bengali, Marathi, Kannada and Malayalam cloud.google.com | Four voices, ashu, neha, shruti and sameer, in Hindi, Gujarati, Marathi, Bengali, Tamil, Telugu, Kannada and Malayalam, plus Hinglish |
| Detail | Google Cloud Text-to-Speech | Voice Infra |
|---|---|---|
| Where it runs | Global, US and EU endpoints. In Asia Pacific, Chirp 3: HD is marked for Tokyo and Singapore the table lists Mumbai with no model marked cloud.google.com | Speech recognition and the model run on our own GPUs in India |
| Detail | Google Cloud Text-to-Speech | Voice Infra |
|---|---|---|
| Try it free | First 1M Chirp 3: HD and Neural2 characters and first 4M WaveNet and Standard characters each month billing must be enabled cloud.google.com | No sign-up to look around the sandbox ₹200 of credit once you confirm a work email; clips you play in the console are not charged |
| How to switch | Google's Cloud Text-to-Speech API and client libraries cloud.google.com | OpenAI's POST /v1/audio/speech: any OpenAI SDK works with our base URL and model mira-tts |
| Detail | Google Cloud Text-to-Speech | Voice Infra |
|---|---|---|
| Streaming | Two-way streaming synthesis, in Preview, for Chirp 3: HD only streaming output is ALAW, MULAW, OGG_OPUS or PCM, and SSML is not supported when streaming cloud.google.com | Raw PCM streams over HTTP as it renders a WebSocket endpoint keeps one connection open for a whole call |
Spot something out of date? Tell us. Every platform in one table: voice AI platforms compared.
Why switch
Everything a phone call in India needs, from one team, priced in rupees.
₹1.80 per 1,000 characters.⁴
On pay as you go and on Business, paid from one rupee wallet. No plan to pick first.
Hindi, Gujarati, Marathi, Bengali, Tamil, Telugu, Kannada and Malayalam. And the Hinglish in between, mid-sentence.
The listening and the thinking: Mira STT and Gemma 4 26B. Mira TTS speaks.
Ask for 8 kHz μ-law or A-law and send it straight down the line. Or take 48 kHz PCM for everything else.
When you want the whole call, the same key and wallet run our voice agents.
AI agents: ₹5/min to start · ₹1.50/min all-in at volume.¹
Real lines from Mira TTS, one request each, unedited. Pick a language and press play.
Hear itA big shelf, or a voice built for the call
Google gives you a big shelf of voices in many Indian languages and a generous free tier, and if you already run on Google Cloud, that's convenient. Our voices are built for the way people speak on calls, slipping into English and back. They cost ₹1.80 per 1,000 characters, billed in rupees.
Google sells speech as part of Google Cloud, with several generations of voices on one price list. For Hindi there are 30 Chirp 3: HD voices, plus older Neural2, WaveNet and Standard ones, and as many Chirp voices for most other Indian languages. There’s also Gemini-TTS, which takes a prompt about how a line should sound.
We do less, on purpose. Our four voices are tuned for one thing: the way people talk on a call in India. Think of “Sir, आपका payment pending है, आज ही कर दीजिए।” Half Hindi, half English, one voice. The Google pages we read don’t mention mixing languages, so test it either way.
Real lines from Mira TTS. Pick one, press play.
Text149 characters
Hi, मैं नेहा बोल रही हूँ Acme Store से। आपका order dispatch हो गया है और Friday तक deliver हो जाएगा। Address में कोई change करना हो तो अभी बता दीजिए।
48 kHz · MonoReady when you are
0:00 / 0:09
Mira TTS, voice neha, rendered 16 Oct 2026. One request, not edited.
Text119 characters
नमस्ते, मैं श्रुति बोल रही हूँ, दाँतों के क्लिनिक से। कल सुबह डॉक्टर साहब से आपका मिलना तय है। क्या आप समय पर आ पाएँगे?
48 kHz · MonoReady when you are
0:00 / 0:08
Mira TTS, voice shruti, rendered 16 Oct 2026. One request, not edited.
Text117 characters
નમસ્તે, હું આશુ બોલું છું. તમારી લોનનો હપ્તો આવતા અઠવાડિયે ભરવાનો છે. તમે કહો તો હું અત્યારે જ ભરવાની રીત સમજાવી દઉં.
48 kHz · MonoReady when you are
0:00 / 0:08
Mira TTS, voice ashu, rendered 17 Oct 2026. One request, not edited.
Text167 characters
வணக்கம், நான் நேஹா பேசுகிறேன். உங்கள் பார்சல் நாளை மாலைக்குள் வந்து சேரும். அந்த நேரத்தில் நீங்கள் வீட்டில் இல்லையென்றால் சொல்லுங்கள், வேறு நாளுக்கு மாற்றி விடுகிறோம்.
48 kHz · MonoReady when you are
0:00 / 0:10
Mira TTS, voice neha, rendered 17 Oct 2026. One request, not edited.
Text114 characters
నమస్కారం, నేను సమీర్ మాట్లాడుతున్నాను. నిన్న మీరు మా కొత్త ఇళ్ళ గురించి అడిగారు. ఇప్పుడు కొంచెం సేపు మాట్లాడవచ్చా?
48 kHz · MonoReady when you are
0:00 / 0:08
Mira TTS, voice sameer, rendered 17 Oct 2026. One request, not edited.
Which to pick
Sometimes it's them. We'll say so.
Google offers global, US and EU endpoints. In its Asia Pacific table, Chirp 3: HD is marked for Tokyo and Singapore, and Mumbai has no model marked. Speech recognition and the model run on our own GPUs in India. If your compliance team needs a precise answer, ask both of us in writing.
Moving
Your code stays. Three values change.
Step 1
Look around with no sign-up. When you are ready, create a key in Developers.
Step 2
The base URL, the model and the voice. The rest of your OpenAI code stays as it is.
Step 3
Four voices, each speaking Hindi, Gujarati, Marathi, Bengali, Tamil, Telugu, Kannada and Malayalam.
The request changes shape. Google’s client takes a voice name and an audio config. Ours is OpenAI’s speech API: a model, a voice, a format and the text. Ask for pcm to stream, ulaw_8000 for a Twilio or Plivo stream, or wav for a file. If you stream LLM text into Google’s two-way synthesis today, our WebSocket endpoint does the same job. Text goes in as the model writes it, and audio comes back a sentence at a time.
Then put two or three Google voices next to ours, reading the lines your product says most. Let people who speak the language pick, without knowing which is which. The Hindi TTS comparison has the rest of the market’s prices.
If the voice is for a calling agent, you may not need to build one. Ours talk in this same voice and cost ₹1.50 a minute all-in on the volume plan.
Python
import os
from openai import OpenAI
client = OpenAI(
api_key=os.environ["MIRA_API_KEY"],
base_url="https://sandbox.voice.miraiminds.co/v1", # 1. our URL
)
with client.audio.speech.with_streaming_response.create(
model="mira-tts", # 2. our model
voice="ashu", # 3. one of our voices
response_format="pcm", # always send it
input="चलिए, आपका order कल deliver होगा।",
) as response:
for chunk in response.iter_bytes():
... # raw 16-bit mono PCM at 48 kHz, playable as it arrives cURL
curl --fail-with-body \
https://sandbox.voice.miraiminds.co/v1/audio/speech \
-H "Authorization: Bearer $MIRA_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "mira-tts",
"voice": "neha",
"response_format": "wav",
"input": "नमस्ते, आपका ऑर्डर तैयार है।"
}' \
--output speech.wav Questions
Sometimes. WaveNet and Standard voices cost ₹380 (US$4) per million characters, and Neural2 ₹1,519 (US$16), both below our ₹1,800. Chirp 3: HD costs ₹2,849 (US$30). And Google's free tier is hard to beat for a prototype.
No. We speak OpenAI's speech API, so you call us with an OpenAI SDK or a plain HTTP request. In most apps that means rewriting one function.
Yes. Ask for ulaw_8000 or alaw_8000 and pass each chunk to your phone provider as it arrives. On a live call, the WebSocket keeps one connection open and lets you stop a reply when the caller cuts in.
Fifteen minutes with the team that builds it. Bring your script, or your hardest call.
Book 15 minutes