Introduction

Udgam Labs just filled out its Shruti on-device speech-to-text family with Shruti Telugu. The product page dates the Telugu drop 4 November 2026, and the Hugging Face card udgamlabs/shruti-telugu landed 2026-11-03. One 29.7 MB W4A8 file (shruti-telugu.cact, ~55.9M params) turns Telugu + English speech into text offline — phone, laptop, or browser WASM — with no server.

Telugu comes out in Telugu script; English loanwords stay in Latin letters the way people actually type (హైదరాబాద్‍లో కొత్త company నుంచి offer letter వచ్చింది). That code-mix behavior is the maker hook.

What Shipped

  • Vendor / product: Shruti Telugu — Udgam Labs (family: Hindi Oct 19, Marathi Nov 2, Tamil/Bengali Nov 3, Telugu Nov 4)
  • Modality: speech-to-text / ASR (on-device)
  • Open vs closed: open weights, Apache-2.0 (built on Cactus Whistle + needle engine)
  • Runnable today: pip install cactus-needle + hf download udgamlabs/shruti-telugu shruti-telugu.cact; CLI ./needle --model shruti-telugu.cact --audio clip.wav; in-browser demo on the product page
  • Concrete fact: on DiarBench code-mix (English words scored in English), Shruti Telugu hits WER 45.4 vs IndicConformer Telugu 52.9 and Whisper-small Telugu FT 55.7 — at roughly 1/5th and 1/32nd those file sizes
  • Device note: ~10× realtime and ~64 MB RAM on a 2021 Galaxy F62 (native engine); 0 words invented on 100 silent clips

Primary page: Shruti | Udgam Labs.

Run It

pip install cactus-needle
hf download udgamlabs/shruti-telugu shruti-telugu.cact config.json --local-dir .

from needle import Whistle

model = Whistle("shruti-telugu.cact")
print(model.transcribe("clip.wav"))  # Telugu / English auto-detect

Set NEEDLE_TELEMETRY=0 if you want the Python package quiet. Native binaries for Mac/Linux/Windows/Android/iOS + WASM live under Cactus-Compute/needle3.

Same-Week Indic Family

This cycle’s zero-day is Telugu, but the same window also shipped siblings that are still missing from artrealmai.com:

Each language is its own download — apps pull only what they need.

Limits Worth Knowing

  • On read Telugu (FLEURS / Kathbath), larger models still win (IndicConformer ~24.5 / 23.0 vs Shruti 38.0 / 32.9).
  • Roughly half of English words still land in Telugu script; no punctuation; numbers come out as words.
  • Noise, music, far-field, and speaker-played audio were not the vendor’s Telugu test focus.

Original Source

Conclusion

If you need privacy-first Telugu+English ASR that fits in an app download and keeps Hinglish-style Latin English inside the transcript, Shruti Telugu is the day-0 open file to try — and the Marathi/Tamil/Bengali siblings make it a real Indic offline kit, not a one-off demo.

—Anabel ♡