.png&w=256&q=75)
🤓 Latest Submissions
RollCaller
A name is one of the most personal things a person carries — tied to family, culture, and identity. Yet in a survey across more than 100 organizations for the "My Name Is" campaign, 73% of people reported having their name mispronounced, many left feeling unimportant and out of place. Research explains why it stings: saying a name correctly fosters belonging, while getting it wrong can induce genuine feelings of alienation. At a ceremony meant to honor you, that gap between recognition and erasure is at its sharpest; That's where RollCaller comes in. (source: https://www.inclusiveemployers.co.uk/blog/understanding-the-importance-of-name-pronunciation/) Participants record themselves saying their own name, and on the day a single, consistent AI voice reads the entire roll back with each name in its owner's actual pronunciation, no presenter stumbling over unfamiliar spellings. We've incorporated thorough a TTS fallback. When someone hasn't recorded themselves, Gemma reads the raw spelling (in any language or script) and reasons out the correct IPA phonetic transcription, a grapheme-to-phoneme task where rule-based tools break down on multilingual and unusual names. Gemma also powers the app's agentic features. wav2vec2 handles audio-to-IPA when a recording exists, eSpeak NG provides a phonemic safety net, and Kokoro-82M synthesizes the final voice. Every model is open — Gemma, wav2vec2, Kokoro-82M, and eSpeak NG — built with GLM-5.2 via Fireworks. Open models, doing dignified work at a moment that matters.
13 Jul 2026
