Ramees Muhammed (RedZoc)
ReportI am a blind software architect in India (backend, cloud and AI) and I use a screen reader every day. RedZoc is my own brand for apps that speak: voice and accessibility apps for blind and low-vision people, made for India and used worldwide. I maintain eSpeak NG Text-to-Speech, an independent Android build of eSpeak NG with over 100,000 installs on Google Play, and I contribute fixes back to the official eSpeak NG project, mostly for Indian languages. Fixes are found and checked with blind users who speak each language.
eSpeak NG for Indian languages
eSpeak NG is an open-source speech engine used by screen readers such as NVDA on Windows and by many Android phones. In Indian languages it still makes mistakes that blind users hear every day: wrong nasal sounds in Hindi, extra or missing vowels, letters and vowel signs read by the wrong name, words spelled out instead of read. This project fixes Indian languages in eSpeak NG together with blind native speakers, and sends every fix with tests to the official eSpeak NG project, so every app and screen reader that uses eSpeak NG benefits. What exists today: fixes sent upstream for Hindi, Marathi, Konkani, Bengali, Assamese, Sinhala, Malayalam text with English words, and more (https://github.com/espeak-ng/espeak-ng/pulls?q=author%3Arameeszone); eSpeak NG Text-to-Speech for Android, over 100,000 installs, whose speech engine source is published for each release (branch redzoc-android, GPL-3.0); and Try eSpeak NG, a web tool that speaks text in over 100 languages in the browser (https://www.redzoc.com/espeak/try/). To be clear: the Android app's screens, ads and forum are not open source. Its speech engine is, and the engine and the language data are what this funding is for. The plan: about 10 to 15 hours a week for 6 months, in three milestones of INR 2,00,000 (the funding plans). Each milestone: INR 1,60,000 for development time and INR 40,000 for blind native-speaker testers. Method for each language: build a list of common words, find where eSpeak NG reads them wrongly, check with blind native speakers, fix the language rules with tests, send the fix upstream, and ship it in the Android app.
Milestone 1 (months 1-2): Hindi and Marathi
Finish the open work in Hindi and Marathi: nasal sounds, extra vowels in words like पहुंचने, and the old exception lists reviewed with blind testers. Get the open pull requests reviewed upstream.
Milestone 2 (months 3-4): Malayalam and Tamil
Malayalam and Tamil, then Telugu and Kannada as time allows: word lists checked with blind native speakers, fixes with tests sent upstream and shipped in the Android app.
Milestone 3 (months 5-6): Bengali, Gujarati, Punjabi, Urdu
Bengali, Gujarati, Punjabi and Urdu, by the same method. Publish the word lists as test data and a short public report of what was fixed and what remains.