Chariot-1-ASR Realtime
Text arrives while people are still talking. On real Indian English phone calls it is level with the best of the Open ASR Leaderboard’s 64 entries, and the same model hears Hindi, Tamil and nine more Indian languages.
Free to start · No card · Live audio or file upload · Indian English
The clip
English words
Clip: FLEURS, CC BY 4.0
#1on real Indian English phone calls, level with the best of the Open ASR Leaderboard’s 64 entries
3.00word error rate on those calls, scored with the leaderboard’s own code
17%fewer errors than OpenAI gpt-transcribe on the same calls
12languages in one model: English and 11 Indian languages, detected from the audio
Benchmark · October 2026
English is not in our ten-language benchmark. This is its result on real calls, against the Open ASR Leaderboard.
Monsoon: unscripted two-person phone calls in Indian English. Word error rate, scored with the leaderboard’s own normaliser and code, lower is better. The other systems’ numbers are the leaderboard’s own; 12 of its 64 entries are shown.
Live audio and recordings
Each session takes a mode, a language or none, and how English words are written. One price for all of it.
Text arrives sooner. For voice agents, live captions and anything a person reacts to as they speak.
The model takes more context before it commits text. For call and meeting transcripts and dictation, where accuracy matters more than a short delay.
Set the language when you know what callers speak, or leave it out and the model detects it from the audio.
Codemix keeps borrowed English words in Latin script. Native writes them in the script of the spoken language.
8 kHz mu-law and A-law straight from a phone line, or 16-bit PCM from 8 to 48 kHz.
Upload a recording in the web app and get its transcript back. WAV, MP3, FLAC, OGG, M4A, AAC or WebM.
API
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=en&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="")Use cases
Hear a caller while they are still speaking, and answer without a pause.
Every English support and sales call written down, straight from the phone line.
Call transcription ›Your data
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.
On Monsoon’s unscripted Indian English phone calls, scored with the Open ASR Leaderboard’s own normaliser and code, it has a 3.00% word error rate: level with the best of the leaderboard’s 64 entries, Microsoft Azure Speech and Zoom Scribe v2 Pro, and ahead of AssemblyAI Universal-3.5 Pro at 3.49% and OpenAI gpt-transcribe at 3.63%.
Yes. Chariot-1-ASR Realtime is built for English and Indian languages, including speech that mixes them in one sentence. Its English result above is on real Indian English phone calls.
Yes. For speech that is mostly Hindi with English words, set language=hi and choose codemix to keep the English words in Latin letters. The Hinglish speech to text page shows both styles.
No. Set language=en when you know the speech is in English, or leave it out and the model detects the language from the audio.
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.
Yes. Stream 16-bit PCM at 8 to 48 kHz, or telephony audio as 8 kHz mu-law or A-law, mono.
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.
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.
In India. Audio you send and its transcripts are processed on Indian infrastructure, under the Digital Personal Data Protection Act.
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.
Start free with 10,000 credits, almost two hours of audio. No credit card.
