Real-Time Call-Center Voice Changer

I need a production-ready, real-time voice-changing application for my call-center environment. Agents must be able to switch among six high-quality U.S.-accented voices—three male and three female—without audible lag while they are on live VoIP calls. Clean, natural-sounding tone modulation is essential; the result should feel like a genuine human voice rather than a gimmick.

The desktop client has to run smoothly on both Windows and macOS and sit between the agent’s headset and any softphone, SIP desk phone, or WebRTC browser tab. Latency must remain low enough that neither party notices a delay, even during high-traffic periods.

An accompanying web-based admin panel lets me create, suspend, and delete user accounts, set seat limits, and pull usage reports by user, date range, and chosen voice. CSV or JSON export for those reports is preferred so I can feed them into our BI tools later.

I am open to whichever DSP, AI, or machine-learning stack you feel delivers the most natural result—whether that’s C++ with a custom neural network, a Rust audio engine, or a Python/TensorFlow backend—as long as deployment for both operating systems is straightforward.

Deliverables
• Agent-side application for Windows & macOS (installer or signed package)
• Web admin portal with user-management and reporting features
• Documentation covering installation, configuration, and an API (if exposed)
• A brief demo session that proves the voices work in real-time on a live SIP or WebRTC call

Acceptance criteria
• Sub-150 ms end-to-end latency during a two-way VoIP call
• Clearly distinguishable U.S. male and female voices with no robotic artifacts
• Successful generation of per-user usage reports from the admin panel

Please include a short outline of your proposed technical approach and any past work that demonstrates real-time audio processing expertise when you reply.