Bangla AI Boost: Govt Nears Completion of Language Tech Project
Spoken Bangla will be transcribed automatically. Written text will be read aloud by computers. Printed books and documents will be converted into digital form within moments. Accurate machine translation of Bangla—without the help of Google—will become a reality. In this way, a massive corpus of Bangla language data, both spoken and written, is being built.
With these ambitions, the “Bangla Language Enrichment through ICT Research and Development Project” has been working since 2016 to develop 16 technological components. The project tenure was extended twice, raising the budget from Tk 159 crore to Tk 855.48 crore. The current phase of the project is scheduled to conclude this December.
Ahead of the project’s tenth anniversary, it has been revealed that several tools have already been developed, including Bangla OCR, speech recognition software, a spell checker, Bangla-to-English translation, text-to-speech, and text transcriber systems. Improved versions of these tools are underway, and a new comprehensive Bangla font named Julai is set to be released on 15 December. A feasibility study is now in progress for a multimodal (speech, text, image) Bangla Large Language Model (LLM) in the project’s next phase. Alongside this, an initiative has been taken to build a national cloud computing facility to support the effort.
These updates were shared by Faiz Ahmad Taiyeb, Special Assistant to the Chief Adviser assigned to the Ministry of Posts, Telecommunications and Information Technology. He said the Julai font ensures precise spacing between Bangla and English characters, including various conjunct letters, vowels, and consonants.
He added that, to enrich Bangla in cyberspace, the project—overseen by the ICT Division and implemented by the Bangladesh Computer Council—has already developed an integrated platform called kagoj.ai. This will serve as a common LLM training and inference platform for the AI-supportive software industry.
A visit to the portal shows six modules that can transcribe Bangla audio, autocorrect misspelled words, translate Bangla to English, and read aloud written text. It can also convert printed Bangla text from images into digital form. However, each feature currently allows only 100–200 words. The platform does not accept text files for uploads, and the voice-to-text tool’s accuracy is still not fully satisfactory. A “feedback” option has been provided for user suggestions.
Authorities, however, remain optimistic. They say the developed Bangla OCR can digitize government and private documents, enable archiving, convert old books and files into digital formats, and support paperless office operations.
Currently, kagoj.ai offers 1,000 free credits per month. One thousand credits allow users to transcribe one hour of speech, or perform 5,000 words of TTS/spell check, or process 50 images/pages through OCR. Additional credits may be requested for research needs.
Faiz Ahmad Taiyeb further stated that APIs will be opened for organizations. According to him, both streaming and non-streaming Bangla speech recognition software have been developed. These tools support speech-to-text and text-to-speech conversions. The speech-to-text system has already been successfully piloted in the government’s electronic filing system, D-Nothi, and is now integrated as a plug-in tool. On Victory Day, the APIs will be opened to all institutions working on Bangla language technology.
DBTech/MUM/IK/OR
