Who's building it

The government-funded model programme at IIT Bombay. Its speech models sit alongside the Param text models.

• Shrutam 2
• Sooktam 2

NoBroker's in-house voice stack, trained for its own customer calls and now sold to other enterprises.

• Akshara STT
• Ragini TTS

AI dubbing and voiceover studio that open-sourced its MahaTTS speech models.

• MahaTTS v2 (Open weights)

Vaanika clones a voice from ten seconds of audio and speaks it in seven Indian languages and English.

• Vaanika

Personalised video at scale. Its Myna models give those videos Indian-language voices.

• Myna-Mini

Speech-first since 2016. Trains its own recognition, synthesis and voice-to-voice models for bank and insurance call centres.

• Vachana STT
• Vachana TTS
• Inya VoiceOS

Long-running Indian speech engine for IVR systems, station announcements and screen readers.

Speech-data arm of the vernacular media company. Released Human-1, the first open full-duplex conversation model in Hindi.

• Human-1 (Open weights)

Open Indic text-to-speech with Svara, plus licensed Indian speech data for other model builders.

• Svara TTS v1 (Open weights)
• Svara Turbo

Two founders building expressive voices for Indian developers. Veena speaks Hindi and English; both open models are Apache 2.0.

• Veena (Open weights)
• Maya1 (Open weights)

Bootstrapped team training its own Hindi-first speech models and running the calling stack in India.

• Mira STT
• Mira TTS

Bootstrapped team building one voice-to-voice model for sales calls, in place of the STT, LLM and TTS chain.

Voiceover studio and TTS API started by three IIT Kharagpur batchmates, with its model team in India.

• Falcon

Speech AI for the Northeast. Lemka listens and speaks in Khasi, Garo, Mizo, Meitei, Kokborok and Nagamese.

• Lemka
• Nyari
• Garo ASR (Open weights)

Its Bodhi speech stack and agent platform are built for Indian banks, lenders and their customers' dialects.

• Bodhi

One of India's oldest language-technology companies: speech recognition and the Gopal voice engine, backed by Reliance Jio.

• Gopal TTS

Research lab behind Ira, an AI companion that phones you and sends voice notes, and the open Rumik-OSS-1 TTS for 22 Indian languages.

• Rumik-OSS-1 (Open weights)

Edge-AI company that published Samba-ASR, a speech recogniser built on state-space models.

• Samba-ASR
• Svara

Speech recognition, speech synthesis and dubbing for all 22 scheduled languages, sold as APIs and its own agent platform.

• Saaras V4
• Bulbul v3
• Sarvam Dub
• Shuka v1 (Open weights)

Speech recognition and synthesis that enterprises can run on their own servers, with voice agents on top.

• Pingala V1 (Open weights)
• Zero TTS

Low-latency speech models and the Atoms agent platform, engineered in Bengaluru for customers worldwide.

• Lightning
• Pulse
• Hydra

Panini TTS speaks 250+ languages, including Indian dialects most models skip: Bhojpuri, Garhwali, Marwari.

• Panini

Small team working on latent speech models, shipping its own TTS for enterprise voice agents.

• Vaani TTS

Bootstrapped studio for emotional text-to-speech in Hindi, Marathi, Tamil and more, for creators and call centres.

Two founders building receptionist agents, with their own Tamil speech model in training.

Long outbound calls for banks and lenders, run on its own domain-tuned language model.

Multilingual voice agents for lenders and financial-services customer care.

Apna's enterprise voice-agent business, live at lenders, insurers and hospitals.

Developer-first voice agents for Indian languages that route every call across the best speech and language models.

Pay-as-you-go phone agents in Hindi, Marathi, Tamil and more, hosted in India.

Bulk AI calling campaigns in Hindi and regional languages, from Dynopii.

AI callers in Hindi, Hinglish and 13 more languages for real estate, D2C brands and SMEs.

Moved from call-quality analytics to AI phone agents for sales, service and payment follow-ups.

Builds BharatGPT assistants, including IRCTC's AskDISHA, which answers train passengers by voice in Indian languages.

Calling agents for sales follow-ups, collections, bookings and recruitment screening.

AI front desk for clinics, dealerships and CAs: answers calls, books slots, collects payments.

Admissions and lead-calling agents for universities, some running 100,000 calls a month.

India's cloud-telephony mainstay, now selling generative voice agents on top of its phone lines.

Voicebots and real-time agent assist for BPOs and contact centres.

Agentic voice and chat for banks and insurers, from one of India's oldest AI startups.

Voice agents for collections, upselling and borrower profiling at banks and NBFCs.

Voice agents and speech analytics for banks, NBFCs, brokers and insurers, in Indian languages.

Contact-centre software whose Hexa voice bot switches between Indian languages mid-conversation.

Jio-owned conversational AI company running voice and chat agents for large enterprises.

Hindi and Hinglish voice receptionists for hotels, clinics and real-estate teams.

Speech technology for in-car voice assistants and contact-centre agents.

Business phone system for Indian SMBs, now with AI voice agents that switch languages mid-call.

Myntra co-founder Mukesh Bansal's enterprise agent company. NuPlay handles voice; Verloop.io now handles chat.

Inbound and outbound calling agents with CRM hooks, from Surat's Krira AI.

Bootstrapped and profitable. Sales, renewal and recovery calls for banks and telcos.

One of India's first cloud contact-centre companies, now adding agentic voice AI for its 3,500+ brands.

Voice API company co-founded by Venky Balasubramanian, now with a platform built for AI voice agents.

Front-desk voice agents for hospitals: appointments, reminders and patient questions around the clock.

Multilingual agents that call back users who drop off mid-KYC and walk them through it.

Contact-centre voice bots and call QA for lenders, with KYC and collections workflows.

Inbound and outbound agents across phone, WhatsApp and web, in more than 20 languages.

YC W26 team whose founders learned the job running a national helpline voice agent across India.

Started as Vernacular.ai in Bengaluru; now runs debt-collection voice agents for lenders.

Outbound sales calling that pairs AI agents with human telecallers, in nine Indian languages.

IIT Bombay founders building an AI BPO: voice agents plus the workflows behind them, for banks and fintechs.

Builds the whole stack in-house, from telephony to a speech recogniser trained on 100,000 hours of Indian calls.

Tests and monitors voice agents: finds where calls fail before customers do.

A phone receptionist for clinics that books appointments in Indian languages and confirms them on WhatsApp.

Self-serve voice agents for lead follow-ups, support calls and first-round interviews.

Voice and WhatsApp agents for collections and citizen services, with data kept in India.

Voice agents for outreach and role-play practice, running its full stack on Indian data centres.

Began at IIT Madras making phones understand Indian languages; grew into a global enterprise AI company.

Enterprise voice agents that take calls in all 22 scheduled languages.

Calling agents for car dealerships: test-drive follow-ups, service reminders, insurance renewals.

Telephony built for voice AI: SIP trunks and real-time call streaming for agent companies.

Outbound voice agents for collections and outreach, built by a Bengaluru team.

No-code voice agents on Indian phone numbers, made for small businesses.

Calling agents for small teams: COD confirmations, lead qualification and admissions outreach.

Enterprise conversation platform born in Bengaluru, with its own voice-cloning model.

• Nexus Vox

Transcribes witness testimony and judges' dictation in Indian courtrooms, in 15 languages, so proceedings stop waiting on stenographers.

Medical speech recognition from a Mumbai transcription company: doctors dictate, notes appear in the record.

EkaScribe listens to crowded, code-mixed OPD consultations and writes up the prescription.

• Parrotlet-a (Open weights)

Voiceover, transcription and voice cloning for Indian-language publishers and creators.

Farmers ask crop questions out loud in Hindi or Hinglish; Dhenu, an agriculture model, answers.

AI spoken-English coach for Hindi speakers who freeze up on calls and in interviews.

Voice-first English tutor that grades pronunciation and fluency as you talk.

Personal call assistant that answers your missed calls in Hindi and Hinglish and blocks spam, on its own speech engine.

Its AI recruiter phones blue-collar applicants, explains the job and screens them for Zomato, Swiggy and Zepto.

Five people building a ledger shopkeepers keep by speaking to it.

IIT Madras lab whose free models and datasets sit underneath much of Indian speech AI.

• IndicConformer (Open weights)
• IndicF5 (Open weights)
• Indic Parler-TTS (Open weights)

The government's language mission: free speech recognition, synthesis and translation APIs for 22 languages.

• VoicERA (Open weights)

Released Vakyansh, an open Indic speech-recognition toolkit, and co-built VoicERA with BHASHINI.

• Vakyansh (Open weights)

Pays rural Indians to record speech in their own languages, then supplies the data to model builders.

Speech and language research, including synthesis for Bhili, Gondi, Santhali and Mundari.

ARTPARK and IISc, with Google, recording how India actually speaks, district by district.

• SraVaani (Open weights)
• DhVaani (Open weights)

Anchored India's TTS consortium for over a decade. Its IndicTTS recordings still train new voice models today.

Speech lab behind RESPIN and SYSPIN, recognition and synthesis data for farmers and low-resource languages.

Recognition and synthesis research; trained Indic-Mio, an open TTS for all 22 languages, on a single GPU in six hours.

• Indic-Mio (Open weights)

Open-source conversation tool for deaf users: simplified live transcripts, and a voice for what they type.

Nonprofit whose Vachan Samiksha listens to children read aloud and scores their fluency for teachers.