1. What inspired you to build this? While exploring voice AI for regional Indian languages, I noted existing tools could synthesise spoken Hindi or Indic speech -- but were purely audio generators. In applications like edtech, digital textbooks, and interactive language readers, audio alone is not enough; the text on the screen also needs to highlight words in real-time, as they're being spoken. Seeing that EdTech builders were forced to spend hours and hours recording and time stamping each little audio segment was an "aha" moment that we needed to build an automated, developer-friendly, regional-indic-specific, time stamping pipeline. 2. What problem were you trying to solve? The problem is that existing Indic TTS engines don't provide word-level timestamps, and standard alignment tools fail on regional languages: The Highlighting Gap: Without millisecond start/end timestamps per word, building interactive, karaoke-style read-along learning experiences are impossible out-of-the-box; The Manual Labor Tax: Teams are forced to either make studio recordings and hire contractors to timestamp every audio segment (dozens of hours of work per course), or they show flat, unsync text; ASR Drift in Indic Languages: trying to use standard speech-to-text (like Whisper) to extract timestamps from generated audio fails often due to phonetic spelling differences and dialect drifts in regional Indian languages, and doesn't match the original written script across 16+ languages (Assamese, Bengali, Sanskrit, Hindi, etc). 3. How did your approach or process evolve while working on the launch? Our process evolved through three phases: Pipeline Evolution (Solving Drift): We saw early that connecting a TTS directly to an ASR alignment tool would produce transcription mismatches against the source text. We engineered a 3-step pipeline for regional synthesis -> sub-word acoustic alignment via Whisper -> an LLM reconciliation layer snapping acoustic timestamps directly back onto the original source script without losing sync. Focusing the ICP (From General Devs to EdTech): While we started with a "developer tool" pitch, we found our highest-value early adopters were not general API consumers, but EdTech innovators, and began shifting our positioning to focus on interactive reading platforms and word-level alignment use-cases; Launch Strategy (Organic & Technical over Paid Hype): Rather than spend on paid ads, we went fully builder-first in the launch: sharing technical deep dives on the engineering stack, giving out free developer sandbox tokens, and demonstrating timestamped JSON payloads alongside audio highlighting in action.

VaaniTTS

Speak Every Indian Voice

Speak Every Indian Voice