VoxIntel

We built real time voice moderation for games that detects emotion and toxicity, makes quick ban decisions, and provides with full visibility through live dashboards.

Comment

Real Time Voice Moderation for Online Games

Inspiration

Online multiplayer games are intense, emotional, and competitive. Voice chat makes them immersive, but it also opens the door to toxic behavior that ruins the experience for others.

Most moderation systems are reactive. Players report abuse after damage is done. We wanted to build something proactive. Something that listens, understands emotional context, measures toxicity, and takes action instantly.

Our goal was simple: protect communities in real time, not after the fact.

What it does

We created an end to end workflow using Airia.ai that:

  1. Transcribes live voice chat using the Modulate API
  2. Detects emotional tone from speech
  3. Generates a toxicity score
  4. Applies decision logic to warn or ban players
  5. Sends structured moderation reports to Lightdash for analytics

The system converts voice into measurable signals.

We treat toxicity as a structured signal:

$$ T = f(E, C, S) $$

Where:

  • (E) = emotional intensity
  • (C) = contextual language toxicity
  • (S) = system confidence

If the final toxicity score crosses a defined threshold:

$$ \text{Ban if } T > \tau $$

The decision is automatic, transparent, and logged.

How we built it

We designed an AI workflow on Airia.ai that orchestrates:

Voice Input → Transcription → Emotion Detection → Toxicity Scoring → Decision Engine → Analytics Output

The moderation output is stored in a structured format and visualized using Lightdash dashboards. This allows studios to monitor:

  • Toxicity distribution
  • Emotional spikes
  • Confidence levels
  • Ban frequency trends

Instead of blind enforcement, studios now have visibility.

Challenges we ran into

1. Voice Context Is Complex

Tone matters. Emotion can amplify toxicity or completely change intent. Balancing emotional intensity with textual meaning was critical.

2. False Positives

An overly aggressive threshold risks banning passionate but harmless players. We had to carefully tune decision thresholds to minimize unfair bans.

3. Real Time Processing

Latency matters in gaming. The system needed to analyze and decide quickly enough to act during live gameplay.

4. Transparency

Automated bans without explanation erode trust. That is why every decision generates a structured report.

Accomplishments that we're proud of

  • Emotion detection significantly improves toxicity evaluation accuracy.
  • Real time moderation must balance fairness and safety.
  • Structured analytics builds trust in automated systems.
  • Community safety is not just detection. It is visibility and accountability.

What we learned

  • Emotion detection significantly improves toxicity evaluation accuracy.
  • Real time moderation must balance fairness and safety.
  • Structured analytics builds trust in automated systems.
  • Community safety is not just detection. It is visibility and accountability.

What's next for VoxIntel

This project is more than a moderation tool.

It is a foundation for intelligent community safety systems that:

  • Protect players in real time
  • Give studios actionable insights
  • Scale across multiplayer ecosystems

We are building infrastructure for safer digital communities.

And we are just getting started.

Built With

Share this project:

Updates