Emilia
Emilia is an ongoing autoresearch project aiming to build a state-of-the-art detector for AI-generated audio.
Core trick
Runs a synthetic-voice detector on live microphone or system audio, feeding its scores and transcript to a scam-context model.
Description
Emilia is an ongoing autoresearch project aiming to build a state-of-the-art detector for AI-generated audio. For the past few weeks, Codex and OpenAI’s latest models have helped direct experiments, evaluate results, and iterate on the detector.
For this hackathon, I built and shipped a native Mac app that brings the current Emilia detector into live conversations. It listens through the microphone or system audio and displays a click-through amber indicator when it detects synthetic-voice evidence. OpenAI Whisper or Realtime transcribes the conversation, and GPT-6 Astra combines the transcript with recent voice evidence to identify scam warning signs. Suspicious context triggers a red border and a warning grounded in the caller’s words.
The distinction matters: an automated appointment reminder can produce amber without red, while a synthetic voice claiming to be your daughter can prompt a warning to verify the caller.
The app, source code, and signed, notarized download are public. Emilia’s research predates the hackathon; the native app and live integrations were built during the event. Because the research uses rights-restricted data, we are not distributing the Emilia checkpoint. The public download instead includes a clearly labeled AASIST-L baseline; the demo uses actual Emilia v8.
README excerpt
# Emilia **A second opinion for what you're hearing.** Watch the one-minute demo below. It uses actual Emilia v8; the public app includes the separately labeled AASIST-L baseline. https://github.com/user-attachments/assets/341f5dcc-5401-40f1-acba-bc40e4085eef **Hackathon judges:** [Watch the 60-second demo](https://github.com/AbregaInc/Emilia-Mac/releases/download/v0.2.0/Emilia-60s-demo.mp4) · [Download the signed Mac app](https://github.com/AbregaInc/Emilia-Mac/releases/download/v0.3.0/Emilia-macOS-arm64.zip) · [Review guide](docs/judges.md) The demo uses the actual Emilia v8 research detector. Emilia was developed using data that we cannot redistribute; this public release excludes its training data, checkpoint and research-only reference audio. The downloadable app includes a separately labeled **NAVER AASIST-L baseline**, running locally through Core ML without Python or another model download. Select it or the external Emilia v8 detector in Settings. The baseline can supply amber and voice-supported warnings; it is not the Emilia model shown in the video. See [baseline provenance and limitations](docs/baseline.md). A native Mac app with Dock and menu-bar controls that listens to a microphone or system audio, transcribes with local Whisper or OpenAI Realtime, and asks GPT-6 Astra to identify actionable scam warning signs. Evidence-backed warnings appear as a red screen border and a movable, nonactivating card. The border does not intercept clicks. ## Build and run