Ceremoni

AI-powered graduation name pronunciation system for universities.

Every graduation, name readers stumble through hundreds of names - especially international, non-Western, or unusually spelled ones. Ceremoni eliminates that. Students record themselves saying their name, the system extracts the exact phonetic pronunciation, and generates ceremony-ready audio that says each name the way the student said it.

Designed to be deployed at any institution - currently in active conversations with Texas Tech University to integrate Ceremoni into the official commencement workflow.

How It Works

Student submits Microsoft Form
  → name, email, college, major, voice recording
      ↓
Backend syncs from OneDrive (Microsoft Graph API)
  → downloads submissions + audio files
      ↓
AI Processing Pipeline
  → audio cleanup (noise reduction, silence trimming)
  → GPT-audio-1.5 listens to recording → IPA transcription
  → Azure Speech Services TTS → ceremony-quality audio
      ↓
Admin Dashboard
  → students organized by session → college → major → name
  → "Play Next" steps through each graduate
  → plays AI-generated pronunciation audio

Pipeline

1. Audio Cleanup — Pydub + noisereduce. Trims silence, normalizes volume, reduces background noise. Accepts any audio format (wav, m4a, mp3, mp4, ogg, flac, etc).

2. Phonetic Extraction — Sends the cleaned audio to OpenAI's gpt-audio-1.5 model. The model listens to the student saying their name and produces an IPA transcription using only Azure-compatible phonemes.

3. Speech Synthesis — Sends the IPA to Azure Speech Services TTS with SSML <phoneme> tags. Generates a clean, ceremony-ready MP3 of the name pronounced correctly.

Tech Stack

• Backend: Python, FastAPI, SQLAlchemy (async), SQLite
• AI: OpenAI GPT-audio-1.5 (phonetic extraction), Azure Speech Services (TTS)
• Audio: Pydub, FFmpeg, noisereduce
• Auth: Microsoft OAuth (MSAL) — works with any institution's Azure AD tenant
• Data Sync: Microsoft Graph API (OneDrive/Forms)
• Frontend: Vanilla JS, SortableJS, Inter font
• Task Queue: Celery + Redis (async processing)

Setup

Prerequisites

• Python 3.12+
• FFmpeg (brew install ffmpeg)
• Redis (for Celery)

Installation

git clone https://github.com/lavneethora/ceremoni.git
cd ceremoni
python -m venv .venv
source .venv/bin/activate
pip install -e .

Configuration

The app requires environment variables for API keys and OAuth credentials. See app/config.py for the full list.

Run

uvicorn app.api.main:app --reload --port 8000

Project Structure

app/
  api/
    main.py              # FastAPI app, lifespan, static files
    routes.py            # Upload/process/audio endpoints
    admin_routes.py      # Auth, sync, ceremony playback endpoints
  models/
    student.py           # Student model
    recording.py         # Recording model (tracks pipeline status)
    ceremony.py          # GraduationEvent, CeremonySession, SessionCollege
  services/
    audio_processor.py   # Noise reduction, trimming, normalization
    phonetic_converter.py # GPT-audio-1.5 → IPA transcription
    tts_generator.py     # Azure TTS with SSML phoneme tags
    pipeline.py          # Orchestrates cleanup → IPA → TTS
    forms_sync.py        # Microsoft Graph API sync from OneDrive
    config_loader.py     # Loads ceremony.yaml into database
    storage.py           # Local file storage abstraction
  templates/
    login.html           # Microsoft OAuth login page
    dashboard.html       # Admin ceremony dashboard
  static/
    dashboard.js         # Session picker, student list, play next
    style.css            # Notion-inspired minimal design
  auth.py                # MSAL OAuth flow
  config.py              # Pydantic settings
  db.py                  # SQLAlchemy async engine + session
ceremony.yaml            # Ceremony setup config

License

Copyright Lavneet Hora 2026
All rights reserved.
This software is not licensed for distribution, modification, or commercial use without explicit written permission from the author.