Hindi phone calls
Error rate with valid spelling variants accepted. Our scores match the Open ASR Leaderboard’s where a model is on both.
Chariot-1-ASR Realtime
Call transcription in the language of the call.
Transcribe support and sales calls in Hindi, English and 10 more Indian languages, live from the phone line or from recordings. On real Hindi phone calls it had the lowest error rate of every system we compared.
Free to start · No card · 8 kHz phone audio · Processed in India
0:00
The clip
English words
Clip: FLEURS, CC BY 4.0
#1on real Hindi phone calls: 5.04, against 8.27 for OpenAI gpt-transcribe
#1on Indian English calls, level with the best of the Open ASR Leaderboard’s 64 entries
8 kHzmu-law and A-law, straight from the phone line
100recordings in a batch through the async API, with a webhook when they finish
Benchmark · October 2026
#1 on real phone calls. In Hindi and Indian English.
Monsoon is a public set of unscripted two-person phone calls in Hindi and Indian English. Lower is better.
Indian English phone calls
Word error rate, scored with the Open ASR Leaderboard’s own normaliser and code. 12 of its 64 entries are shown.
Live or recorded
Every call, during it or after.
Live, during the call
Stream the call’s audio over the WebSocket as 8 kHz mu-law or A-law, and read the transcript while the call is going.
Recordings, after the call
Send up to 100 recordings in one batch, by link or upload, each up to three hours. A webhook tells your system when they are done.
In the web app
Upload a recording in Speech to Text and read its transcript in your browser.
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"
"?language=hi&audio_format=ulaw_8000&mode=high_accuracy")
headers = {"chariotai-api-key": API_KEY}
async with websockets.connect(url, additional_headers=headers) as ws:
await ws.recv() # session.started
for chunk in phone_line(): # 8 kHz mu-law, as the call happens
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="")For support and sales teams
The settings for calls.
The caller’s language
Set the language for a Hindi line, or leave it out and the model detects each caller’s language from the audio.
English words
Codemix keeps borrowed English words in Latin letters; native writes them in the script of the spoken language.
High accuracy
The mode for call transcripts, where accuracy matters more than a short delay: it takes more context before it commits text.
Word timings
Ask for timestamps and each word comes back with its time in the call.
Processed in India
Audio and transcripts are processed on Indian infrastructure, under the Digital Personal Data Protection Act.
Billed by the second
1.5 credits per second of audio, in either mode.
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.
Use cases
Other things people build with it.
Voice agents, too: they hear a caller while they are still speaking, in their language.
Audio and video to text
Upload a recording or a video and get its transcript, in the language it was spoken.
Audio and video to text ›Voice typing
Speak into the microphone, watch the words appear, and copy them anywhere.
Voice typing ›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.
How accurate is it on phone calls?
On Monsoon’s unscripted Hindi phone calls it scored 5.04, the lowest of every system we compared; OpenAI gpt-transcribe scored 8.27 and ElevenLabs Scribe v2 8.45. On the Indian English calls it scored 3.00, level with the best of the Open ASR Leaderboard’s 64 entries.
Can it transcribe calls live?
Yes. Stream the call over the WebSocket as 8 kHz mu-law or A-law, or as 16-bit PCM from 8 to 48 kHz, and the text arrives while the call is going.
Can I send call recordings in bulk?
Yes. The async API takes up to 100 recordings in a batch, by link or upload, each up to three hours, and can call a webhook when they finish.
Does it separate the agent and the customer?
Not yet. Transcripts are plain text, with optional word timestamps.
Where is call audio processed?
In India. Audio you send and its transcripts are processed on Indian infrastructure, under the Digital Personal Data Protection Act.
How is it billed?
1.5 credits per second of audio, 90 credits a minute, from the same credits as text to speech. Every account starts with 10,000 free credits, almost two hours of audio.
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.
