BengaliStoryTeller: A Free, Offline AI That Reads Bengali Bedtime Stories

A developer built BengaliStoryTeller, a free, fully offline text-to-speech tool that reads Bengali bedtime stories aloud sentence by sentence on a CPU-only Ubuntu laptop, inserting natural pauses at the Bengali danda (।) to mimic a human storyteller. The project uses Meta's open-weight facebook/mms-tts-ben model (~145 MB) via Python, Transformers, PyTorch and soundfile, chosen over a larger ~900 MB Coqui Bengali model for its sound, and requires no internet, GPU or subscription. The developer said the tool was inspired by a childhood memory of a sister asking their exhausted parents for bhuter golpo and raja-rani stories each night.

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 4: Ghost in the Machine https://dev.to/challenges/hacktoberfest-week4-2026-10-26 BengaliStoryTeller is a free, offline AI storyteller that reads Bengali stories aloud to children in a calm, natural voice, right from a normal laptop. The problem: After a long, tiring day at work, many parents still hear the same request at bedtime: "Tell me a story " But they're exhausted, short on time, and have often already told every story they know. Bengali-speaking families also have very few storytelling or AI voice tools in their own language , since most are built for English. How it helps: BengaliStoryTeller takes a Bengali story and reads it aloud sentence by sentence, with natural pauses like a real storyteller. Parents get a helping hand on busy or tiring days, and children still get their bedtime story in their mother tongue. It needs no internet, no GPU, and no subscription. 🎧 Listen to a sample story Bengali https://github.com/abhisek247767/BengaliStoryTeller/blob/main/stories/story 3.wav You come home tired after a long day at work, and your child asks, "Tell me a story " Most nights you're too tired to make up something new, and the same old stories get boring. Bengali Story Teller fixes that. Run one command and it: The stories are scary enough to be fun, and every one… BengaliStoryTeller runs fully offline on a CPU-only Ubuntu laptop, with no cloud APIs and no GPU. Model: Meta's open-weight facebook/mms-tts-ben MMS text-to-speech for Bengali from Hugging Face. It's small ~145 MB and fast on a normal CPU. I first tried a larger Coqui Bengali model ~900 MB but preferred how MMS sounded. Stack: Python, Transformers, PyTorch CPU , NumPy, and soundfile. Without open models, this project wouldn't exist. Most closed TTS APIs offer limited Bengali support, charge per character, and need the internet, which is a poor fit for a bedtime story tool used every night. Because MMS is open-weight, I could: Meta trained MMS on over 1,000 languages so that smaller language communities aren't left behind. Open innovation let me take that work and turn it into something useful for Bengali-speaking kids and their parents. This project started at home. When we were kids, my elder sister would go to our parents every single night and ask for a story: sometimes a bhuter golpo ghost story , sometimes a story about raja and rani king and queen . But after a whole day of work, our parents were often just too tired to tell one. That memory stayed with me. Those stories are how so many Bengali children grow up: scared of ghosts, dreaming of kings and queens, and quietly learning their language along the way. The first time my laptop read a full Bengali story out loud, pausing after every । like a real storyteller, I thought of my sister waiting for her story. BengaliStoryTeller is for every child like her, and every tired parent like mine, so the bedtime story never has to be skipped.