From Hindi and Tamil to Bhojpuri, Maithili, and Chhattisgarhi, Horizon Indic ispurpose-built for India's multilingual conversations and mixed-languagespeech. Production-ready speech recognition designed for contactcenters, banking, healthcare, and government applications.
Language Coverage
Most speech recognition systems stop at official languages. Real conversations don't. Customers move between Hindi and English. Families speak Bhojpuri at home but Hindi at work. Government services receive calls in dialects that many AI systems have never evaluated.
Horizon Indic was built for these conversations - combining state-of-the-art recognition across India's scheduled languages with industry-leading support for regional dialects and multilingual speech.
Lower Error Rates Across Every Major Benchmark. Hindi WER 11.9% and It Doesn't Stop at "Hindi".
| System | Kathbath | KB Hard | Common Voice | FLEURS | IndicTTS | RESPIN | Gramvaani | Avg |
|---|---|---|---|---|---|---|---|---|
| Google Cloud STT | 14.30 | 16.70 | 20.80 | 19.40 | 18.30 | n/a | 59.90 | 24.90 |
| AI4Bharat IndicWav2Vec | 12.20 | 16.20 | 20.20 | 18.30 | 15.00 | n/a | 42.10 | 20.70 |
| Microsoft Azure STT | 13.60 | 15.10 | 14.60 | 24.30 | 15.20 | n/a | 42.30 | 20.80 |
| AI4Bharat IndicWhisper | 10.30 | 12.00 | 15.00 | 11.40 | 7.60 | n/a | 26.80 | 13.80 |
| Shunya HEEP-Indic (Horizon Indic) | 8.53 | 8.97 | 9.96 | 11.04 | 6.59 | 12.05 | 25.98 | 11.90 |
14% relative gain over IndicWhisper, 40% relative gain over the next strongest commercial system. First-ever ASR coverage on Bhojpuri, Chhattisgarhi, Magahi, and Maithili (RESPIN benchmark) - dialects that don't appear on any competitor's leaderboard because no competitor has evaluated on them.
Same accuracy across both modes.
Separate speakers even in multilingual conversations.
Extract customer intent alongside transcription.
Hindi-English, Tamil-English, Bengali-English, Marathi-English and more.
Customer names, Village names, Government schemes, Addresses, Organizations, Brand names
Built for phone calls, roadside conversations, and low-bandwidth audio.
