Against models that hear the whole recording
Word error rate, average of 10 Indian languages, lower is better. Every system that takes a language setting was given it.
9%fewer errors than the best batch model, while streaming
Chariot-1-ASR Realtime
Real-time speech to text for Indian languages.
Text arrives while people are still talking. It is #1 in all 10 Indian languages we tested, ahead of every streaming API and even the models that hear the whole recording before they write a word.
Free to start · No card · Live audio or file upload · 12 languages
0:00
The clip
English words
Clips: FLEURS, CC BY 4.0
#1in all 10 Indian languages, against every streaming API and every batch model we tested
#1on real phone calls in Hindi and in Indian English
17%fewer errors than the next best streaming API, without being told the language
26%fewer errors than Google Gemini 3.5 Transcribe, which hears the whole recording first
Benchmark · October 2026
#1 in all 10 Indian languages. Against every system we tested.
79 public test sets from 11 corpora, about 153,000 utterances: read, spontaneous and conversational speech, phone audio and noisy recordings. One scoring code for every system.
Against streaming APIs
Word error rate by language, lower is better. Chariot was never told the language. Next best is the best of the other streaming APIs in that language.
| Language | Chariot | Next best | Fewer errors |
|---|---|---|---|
| Hindi | 12.90 | 16.76 | 23% |
| Marathi | 19.86 | 20.99 | 5% |
| Bengali | 16.43 | 22.21 | 26% |
| Tamil | 24.09 | 30.13 | 20% |
| Telugu | 20.80 | 22.86 | 9% |
| Kannada | 24.21 | 29.86 | 19% |
| Malayalam | 31.24 | 35.70 | 12% |
| Gujarati | 22.35 | 26.59 | 16% |
| Punjabi | 20.49 | 25.03 | 18% |
| Odia | 21.97 | 28.02 | 22% |
#1 on real phone calls, in Hindi and Indian English
Monsoon: unscripted two-person phone calls. Error rates, lower is better.
#1Hindi calls. 5.04, the lowest of every system we compared. OpenAI gpt-transcribe: 8.27. ElevenLabs Scribe v2: 8.45.
#1Indian English calls. 3.00, level with the best of the Open ASR Leaderboard’s 64 entries, scored with the leaderboard’s own code.
Live audio and recordings
Set it once per session. Then talk.
Each session takes a mode, a language or none, and how English words are written. One price for all of it.
Balanced
Text arrives sooner. For voice agents, live captions and anything a person reacts to as they speak.
High accuracy
The model takes more context before it commits text. For call and meeting transcripts and dictation, where accuracy matters more than a short delay.
Any language, or the one you set
Set the language when you know what callers speak, or leave it out and the model detects it from the audio.
Codemix or native
Codemix keeps borrowed English words in Latin script. Native writes them in the script of the spoken language.
Phone audio
8 kHz mu-law and A-law straight from a phone line, or 16-bit PCM from 8 to 48 kHz.
Recordings too
Upload a recording in the web app and get its transcript back. WAV, MP3, FLAC, OGG, M4A, AAC or WebM.
Use cases
What people build with it.
Each has its own page: how it works, the settings for it and what it costs.
Voice agents
Hear a caller while they are still speaking, in their language, and answer without a pause.
Recordings and videos
Upload a recording or a video and get its transcript, in the language it was spoken.
Audio and video to text ›Call transcripts
Every support and sales call written down, in the language it was spoken.
Call transcription ›Voice typing and notes
Notes, forms and messages spoken in Hindi, Tamil or any of the 12 languages.
Voice typing ›API
From audio to text in a few lines.
One WebSocket. Send audio as you capture it, read text as it is recognised, and send input.done for the complete transcript.
# pip install websockets
import json, websockets
url = ("wss://api.chariot.in/v1/asr/ws"
"?audio_format=pcm_16000&mode=balanced")
headers = {"chariotai-api-key": API_KEY}
async with websockets.connect(url, additional_headers=headers) as ws:
await ws.recv() # session.started
for chunk in microphone(): # 100 ms of audio at a time
await ws.send(chunk)
await ws.send(json.dumps({"type": "input.done"}))
async for message in ws:
event = json.loads(message)
if event["type"] == "transcript.delta":
print(event["text"], end="")Pricing
Plans from ₹499 a month.Plans from $5 a month. Or start free.
Speech to text uses the same credits as text to speech, at one rate in both modes. Each plan covers this much audio a month.
| Plan | A month | Speech to text |
|---|---|---|
| Free10,000 credits | ₹0$0 | about 2 hours |
| Starter100,000 credits | ₹499$5 | about 19 hours |
| Creator250,000 credits | ₹999 | about 46 hours |
| Startup1,500,000 credits | ₹4,999$49 | about 280 hours |
| Scale10,000,000 credits | ₹29,999$299 | about 1,900 hours |
Monthly prices. Prices exclude applicable taxes.
Your data
Your audio stays in India.
Speech to text runs on Indian infrastructure. Audio you send and its transcripts are processed in India under the Digital Personal Data Protection Act. Chariot is one of four teams selected under the IndiaAI Mission to build the country’s foundation models.
IndiaAudio and transcripts are processed in India.
DPDPHandled under India’s data protection law.
1 of 4Teams selected under the IndiaAI Mission to build foundation models.
Which languages does it support?
Hindi, Marathi, Bengali, Tamil, Telugu, Kannada, Malayalam, Gujarati, Punjabi, Odia, Sanskrit and English. Set the language for a session, or leave it out and the model detects it from the audio.
Does it handle Hindi mixed with English?
Yes. Speech that mixes a language with English is normal input. Choose how the English words are written: codemix keeps them in Latin script, native writes them in the script of the spoken language.
What is the difference between balanced and high accuracy?
Balanced sends text sooner, for voice agents and live captions. High accuracy takes more context before it commits text, which usually gives a more accurate transcript, for calls, meetings and dictation. Both cost the same.
Can I send phone-call audio?
Yes. Stream 16-bit PCM at 8 to 48 kHz, or telephony audio as 8 kHz mu-law or A-law, mono.
How is it billed?
1.5 credits per second of audio you send, in either mode, from the same credits as text to speech. Time with the connection open but no audio sent is not billed. Every session is saved to your workspace with its transcript and cost.
Can I transcribe a recording?
Yes. Upload it in Speech to Text in the web app: WAV, MP3, FLAC, OGG, M4A, AAC or WebM, up to 3 minutes and 20 MB a file. For a longer recording, stream it over the WebSocket, as the quickstart does: a session runs for up to three hours.
Where is my audio processed?
In India. Audio you send and its transcripts are processed on Indian infrastructure, under the Digital Personal Data Protection Act.
