Canto101: Learn Cantonese with Karaoke & Interactive Role-Play.

Learn Cantonese with AI-powered role-play, karaoke, and games. Discover how to prompt for natural dialogue and use cultural hooks for effective language retention.

Gemini LLM CC-Canto Zeabur Azure Speech Pronunciation Assessment Azure Speech

Overview

Canto101 is a full-stack web app that makes Cantonese learning fun and culturally immersive. Instead of dry textbooks, users chat with an AI in realistic scenarios like ordering milk tea or hailing a taxi, search for Cantonese songs and tap any lyric to instantly see its Jyutping romanisation and meaning, and play vocabulary games that reinforce memory. The live demo includes the AI conversation engine, the smart dictionary with over a thousand entries, the karaoke lyric viewer, and the challenge zone. Real-time pronunciation scoring is firmly on my roadmap and I will also share my architectural blueprint for how I plan to integrate it using Azure Speech in the near future.

Links

https://cantonese101.zeabur.app/
Canto101 teaches Cantonese pronunciation, mapping nine tones to Solfege and pitch contours.

Tech stack

Gemini LLM

Google's native multimodal model family built for cross-reasoning across text, code, images, audio, and video.

Gemini is a frontier-scale model series built from the ground up for native multimodality. It operates across three primary sizes: Ultra for complex reasoning, Pro for scaled performance, and Nano for on-device efficiency. The architecture supports a massive 2-million-token context window, enabling the processing of hour-long videos or codebases exceeding 100,000 lines in a single prompt. By integrating MMLU benchmarks and sophisticated cross-modal reasoning, Gemini provides the backbone for the Vertex AI ecosystem and Google's most advanced developer tools.

https://deepmind.google/technologies/gemini/

View projects
CC-Canto

CC-Canto is an open-source, community-driven Cantonese-English dictionary database designed to power modern language learning applications.

Developed by Pleco Software, CC-Canto bridges a critical gap in Chinese lexicography by providing a comprehensive, open-source Cantonese-English dictionary. The database expands on traditional Mandarin-focused resources (like CC-CEDICT) by incorporating over 120,000 entries complete with Jyutping romanization, colloquial Cantonese expressions, and English definitions. Distributed under a Creative Commons Attribution-ShareAlike 3.0 license, it serves as the foundational data layer for developers building language learning apps, translation tools, and digital dictionaries across iOS, Android, and desktop platforms.

https://www.pleco.com/cc-canto/

View projects
Zeabur

Zeabur is an AI-powered cloud deployment platform that automates infrastructure, CI/CD, and server management with one click.

Zeabur operates as an AI DevOps engineer for modern developers, automating the entire deployment pipeline from GitHub push to live production. The platform automatically detects frameworks (including Node.js, Python, and Go), configures CI/CD, and provisions private networking for databases and services within a single project. Builders can manage custom domains, access an integrated AI Hub for LLM APIs, and deploy pre-built templates like WordPress or n8n instantly. By offering predictable, fixed-rate pricing plans, Zeabur eliminates the risk of surprise billing while giving teams full control over their own servers.

https://zeabur.com

View projects
Azure Speech Pronunciation Assessment

An AI-driven speech evaluation tool that provides real-time, actionable feedback on pronunciation accuracy, fluency, and prosody for language learners.

Azure Speech Pronunciation Assessment acts as a personal, scalable language coach by analyzing spoken audio against native-speaker models (trained on over 100,000 hours of speech data). It delivers immediate, objective scores across key metrics: accuracy, fluency, completeness, and prosody (rhythm and intonation). Developers can integrate this API into educational apps, gaming platforms, or corporate training systems to pinpoint specific mispronunciations down to the phoneme level. With support for both scripted reading and unscripted speaking scenarios, it provides the precise feedback loop learners need to speak with confidence.

https://learn.microsoft.com/en-us/azure/ai-services/speech-service/pronunciation-assessment

View projects
Azure Speech

Canto101: Learn Cantonese with Karaoke & Interactive Role-Play.

Related talks