22 Indian languages natively trained
Hindi, Tamil, Telugu, Bengali, Marathi, Kannada, Malayalam, Gujarati, Punjabi, Odia, Assamese, Urdu + 10 more, plus English (Indian, US, UK accents). Not translated from English: trained on Indian speech.
Speech-to-Text API
Speech-to-text API built for Indian languages
Speech-to-text for 22 Indian languages plus English. Real-time WebSocket streaming, OpenAI-compatible batch, $0.005 per minute. Transparent pricing with no seat fees or minimums.
- 22 Indian languages trained natively (not translated from English)
- Hinglish + Tanglish code-switching handled
- Streaming + batch (file upload)
- $0.005/min, transparent pricing, no seat fees
- Indian languages
- 22
- Real-time transcripts
- Streaming
- Hinglish native
- Code-mixed
- per minute
- $0.005
How it works
Transcribing audio in 4 lines of code
OpenAI-compatible batch. Native WebSocket streaming. Pick your path.
- 01
Grab an API key
Sign up at console.callmissed.com and copy your cm_ API key. 1000 free API credits attached.
- 02
Send audio
POST a WAV, MP3, FLAC, OGG, or M4A file, or stream raw PCM over WebSocket for live audio.
- 03
Get transcript
JSON response with timestamps, speaker diarization, word-level confidence, and automatic language detection.
- 04
Use downstream
Pipe into an LLM for Q&A, summarize meetings, caption videos, or feed your ASR-dependent workflow.
Features
What makes CallMissed STT different
If you've tried Google STT or Whisper on Hindi audio, you know what's wrong. This fixes it.
Streaming + batch modes
WebSocket streaming with partial transcripts for real-time apps, or HTTP batch for files. Same API shape, same model quality.
Hinglish + code-switching
Customers switch between Hindi and English mid-sentence. Our models decode both, tag each word with its source language, and never drop accuracy on mixed speech.
Word-level timestamps + diarization
Every word comes with start/end time and confidence score. Multi-speaker audio gets speaker_0/speaker_1 labels automatically.
Custom vocabulary
Boost recognition of your brand names, product SKUs, medical terms, or regional place names by passing a custom vocab list at request time.
Data privacy + PII redaction
Audio stored in our India region (Azure Central India), auto-deleted after 30 days. Optional on-request PII redaction (phone numbers, Aadhaar, card numbers) at the word level.
Use Cases
Speech-to-text use cases
Real deployments, real data rates, real cost savings.
Call center QA
Transcribe 100% of calls for compliance + coaching
Your call center records 10,000 calls a day. Human QA samples maybe 200 of those. Use the STT API to transcribe all 10,000, then run compliance checks, keyword spotting, and sentiment analysis downstream.
Result
Typically 100% call QA at 1% of manual cost.
Meeting transcription
Indian-language meeting notes + action items
Teams run meetings in Hindi, Tamil, Marathi, or Hinglish. Stream the meeting audio into the STT API, then pipe the transcript to an LLM for summary, action items, and decision log.
Result
Meeting notes in 30 seconds, in the language spoken.
Media & OTT
Indian-language captions and subtitles
Streaming platforms and broadcasters need captions in 10+ Indian languages. The STT API generates timed captions with word-level timestamps, ready for SRT/WebVTT export.
Result
Typically cheaper than human captioning services at this volume.
EdTech
Pronunciation scoring for language learners
Students speak into the app, STT transcribes with confidence per-word, and your app compares expected vs actual phonemes to give pronunciation feedback, in 22 Indian languages.
Result
Built-in pronunciation coach without external vendors.
Field / rural applications
Voice-first data collection
Enumerators in rural India speak survey responses directly into your mobile app (no typing). STT transcribes their speech in the local language, your app extracts structured fields, and the data syncs when connectivity returns.
Result
Data collection in the respondent's language, typically 3x faster.
Accessibility
Real-time captions for live events
Live captions for conferences, webinars, or government broadcasts. Stream audio into the WebSocket API and render partial transcripts in real time, so deaf and hard-of-hearing attendees can follow along.
Result
WCAG-compliant live captions without a human captioner.
Compare
CallMissed STT vs Google, AWS, Whisper, Deepgram
On Indian-language accuracy, we beat every major provider, at a quarter of the price.
| Feature | CallMissed | Google STT | AWS Transcribe | Whisper API | Deepgram |
|---|---|---|---|---|---|
| 22 Indian languages | |||||
| Hinglish + code-switching | |||||
| Streaming WebSocket transcripts | |||||
| Word-level timestamps | |||||
| Speaker diarization | |||||
| Custom vocabulary | |||||
| OpenAI-compatible API shape | |||||
| India data residency | |||||
| Price / minute | $0.005 | $0.016 | $0.024 | $0.006 | $0.0043 |
Comparison based on publicly listed features as of 2026. Check each vendor's site for the latest.
Code
Python, Node, curl: start transcribing in 30 seconds
The CallMissed STT API follows the OpenAI audio transcription shape. If you've ever called openai.audio.transcriptions.create(), this is a drop-in.
transcribe.pyPython
from openai import OpenAI
client = OpenAI(
base_url="https://api.callmissed.com/v1",
api_key="cm_your_key",
)
with open("call_recording.mp3", "rb") as audio:
result = client.audio.transcriptions.create(
model="saaras:v3", # Sarvam — Indian + code-mixed.
# Also: "whisper-large-v3-turbo" (99 langs)
# Also: "nova-3" (Deepgram, diarize + smart-format)
file=audio,
language="hi-IN", # Sarvam BCP-47; "unknown" for auto-detect
response_format="verbose_json",
)
print(result.text)Python: file transcription with speaker diarization
stream.mjsJavaScript
import { CallMissed } from "callmissed";
const cm = new CallMissed({ apiKey: process.env.CM_KEY });
const stream = cm.audio.stt.stream({
model: "saarika-v2",
language: "hi",
sampleRate: 16000,
});
stream.on("partial", (t) => console.log("partial:", t.text));
stream.on("final", (t) => console.log("final:", t.text, "speaker:", t.speaker));
// pipe raw PCM from your mic / getUserMedia
micStream.on("data", (chunk) => stream.send(chunk));Node/JS: streaming transcription from a microphone
FAQ
Speech-to-text API questions, answered
What is a speech-to-text API?
A speech-to-text (STT) API is a REST or WebSocket endpoint that converts audio (microphone, phone call, recorded file) into text transcripts. You send the audio bytes, it returns words with timestamps, confidence scores, and optional speaker labels. CallMissed's STT API is built on models trained for Indian languages and accents, a strong fit for Hindi, Tamil, Telugu, Bengali, Marathi, and 17 more.
Which Indian languages does CallMissed STT support?
All 22 scheduled languages of India: Hindi, Tamil, Telugu, Bengali, Marathi, Kannada, Malayalam, Gujarati, Punjabi, Odia, Assamese, Urdu, Sanskrit, Kashmiri, Konkani, Sindhi, Dogri, Manipuri, Bodo, Maithili, Santali, and Nepali. Plus Indian English, US English, UK English, and 50+ international languages. The models are trained on native Indian speech, not translated from English, so accents, regional dialects, and code-switching are handled natively.
What's the latency on streaming speech-to-text?
Over WebSocket you get partial transcripts in real time as audio arrives, with final transcripts (finalized punctuation and speaker labels) shortly after speech ends. It's built for live captions, voice agents, and real-time voice interfaces. Batch file transcription returns the full transcript in one response.
Can the API handle Hinglish and code-switching?
Yes, and this is where CallMissed stands apart. Our models handle code-switched Indian speech: a customer can say 'Please check mera order status' and we transcribe it correctly, tagging the source language. Sarvam's Saaras models are built for exactly this Indic code-mixing.
How much does speech-to-text cost?
CallMissed STT (Sarvam Saaras) is priced at $0.005 per audio-minute, cheaper than Google STT ($0.016) or AWS Transcribe ($0.024). The free tier includes 1000 API credits. There are no setup fees, minimum commitments, or per-channel charges.
What audio formats and sample rates are supported?
Batch mode: WAV, MP3, FLAC, OGG, M4A, OPUS. Streaming mode: raw PCM (16kHz or 24kHz, 16-bit mono), OPUS, or G.711. Max file size is 100MB for batch; unlimited for streaming. The service auto-detects sample rate from file headers. For phone-audio use cases (8kHz), enable the telephony model for better accuracy on narrow-band speech.
Does it do speaker diarization?
Yes, enable speaker_diarization: true in the request. You'll get speaker_0, speaker_1, etc. labels on each word or utterance. Great for multi-person meetings, call-center calls (customer + agent), or podcasts. Up to 6 speakers handled automatically; more on request.
Can I add custom vocabulary?
Yes. Pass a custom_vocabulary list of up to 1000 words at request time: brand names, product SKUs, medical terms, regional place names, technical jargon. The model boosts recognition of those terms without needing retraining. No extra cost.
Where is audio data stored?
Audio and transcripts are stored in our India region (Azure Central India) and auto-deleted after 30 days by default. You can request shorter retention (7 days) or immediate deletion on Pro/Enterprise plans. CallMissed runs DPDP Act-aligned infrastructure with India data residency. We never use your audio to train shared models.
How do I use it with Python or Node?
The STT API is OpenAI-compatible for batch transcription, so the OpenAI SDK works out of the box. For streaming, use our WebSocket endpoint or the official callmissed-py / callmissed-node SDK. Full docs with copy-paste examples at docs.callmissed.com/stt.
Start transcribing in Hindi in 30 seconds
Sign up free, grab an API key, paste our snippet. 1000 API credits included, about 2,500 minutes of transcription.