Documentation

Complete reference guide for Virtual Video Director — from initial setup through advanced broadcast integration. VVD automates multi-camera switching by detecting who’s speaking and triggering your production infrastructure in real time.

Getting Started

VVD operates on a simple principle: one mic = one camera. Each VVD Channel represents one speaker or audio source. When VVD detects that a speaker’s microphone is active, it fires the triggers assigned to that channel — switching cameras, recalling PTZ presets, or sending commands to your production infrastructure.

Core Concepts

• Channels: Each channel corresponds to one audio source (microphone). You can have 4 to 128 channels depending on your licence tier.
• Trigger Slots: Each channel has 8 trigger slots that fire simultaneously when that channel’s speaker is detected. This allows you to switch a video source, recall a PTZ preset, and send a tally command all at once.
• Special Channels:
  • Channel 129 — Overview: Fires when multiple speakers are talking at the same time (e.g., a wide shot).
  • Channel 130 — Silence: Fires when no one is speaking (e.g., a fallback or beauty shot).
  • Channel 131 — Fixed Timer: Fires at a set interval regardless of audio activity (e.g., periodic establishing shots).

Basic Workflow

1. Configure audio interfaces — Add one or more audio sources (WASAPI, ASIO, NDI, etc.) in the Audio Setup panel.
2. Map channels — Assign physical audio inputs to VVD channels so each microphone feeds its own channel.
3. Set up triggers — For each channel, configure what should happen when that speaker is detected (switch a camera input, recall a PTZ preset, send an HTTP command, etc.).
4. Adjust detection settings — Set attack, decay, gain, and gate for each channel to match your environment.
5. Test — Use the built-in meters and trigger indicators to verify everything is working correctly.
6. Go live — Lock the control panel (padlock icon) to prevent accidental changes, then let VVD handle the switching.

Speech Detection

VVD uses two proprietary detection engines. You can select the mode per channel by right-clicking the channel’s meter display.

Mills Smart Sensor (Recommended)

An AI-powered neural network voice activity detector that runs entirely on-device. It uses the Silero VAD v5 model in ONNX format to distinguish human speech from background noise, music, and other non-speech audio. No cloud connection required — all processing happens locally.

• Processes audio at 16 kHz in 32 ms chunks
• Maintains per-channel state tracking for consistent detection
• Supports DirectML GPU acceleration (NVIDIA, AMD, Intel) with automatic CPU fallback
• Scales to 40+ channels via multi-threaded worker architecture (6 channels per ONNX session)

Mills Level Sensor (Fallback)

A volume-based detection algorithm that analyses relative differences between all audio signals to determine who is speaking. No calibration required — it adapts automatically to any voice, microphone, and room.

The Mills Level Sensor is required for meter-based audio sources (Wheatstone Blade, ClearOne, vMix Direct) where raw audio samples are not available.

Detection Parameters

• Attack (0–1, default 0.5): Controls how quickly detection responds when speech begins. Lower values make detection more responsive; higher values require more sustained speech before triggering.
• Decay (0–1, default 0.3): Controls how long detection holds after speech stops. Lower values release quickly (snappy cuts); higher values hold longer (smoother, more natural switching).

Right-click a channel’s meter to switch between Mills Smart Sensor and Mills Level Sensor for that channel.

Control Panel

The control panel provides global switching parameters that affect how VVD decides when and how to switch between speakers.

Minimum Duration

The lockout period after each switch. During this time, VVD will not switch to a new speaker even if one is detected. This prevents rapid, jarring cuts.

• 1–2 seconds: Fast-paced debates and panel discussions where quick cuts feel natural.
• 2–4 seconds: Normal conversation, interviews, and general use.
• 4–6 seconds: Presentations, lectures, and slower-paced events where stability is preferred.

Overview Mode

Automatically switches to a wide shot (Channel 129) when two or more speakers are talking simultaneously. Configurable parameters include:

• Threshold: How many simultaneous speakers trigger the overview shot.
• Duration: How long the overview shot holds before returning to single-speaker switching.

Silence Detection

Switches to a fallback camera (Channel 130) when no one is speaking. Useful for beauty shots, wide angles, or graphics during pauses.

• Threshold: How long silence must persist before triggering (avoids false triggers during natural speech pauses).
• On Power Off: Option to fire silence triggers when VVD is powered off or in standby.

Realism Simulator

Adds reaction shots by occasionally cutting to recent speakers who are not currently talking. This creates a more natural, broadcast-style feel by showing listener reactions.

• Window: How far back in time to look for recent speakers to use as reaction shot candidates.
• Frequency: How often reaction shots are inserted into the switching sequence.

Fixed Timer

Fires Channel 131’s triggers at a set interval, regardless of audio activity. Useful for periodic establishing shots, cutaway graphics, or timed sponsor bumpers. Intervals can be set up to 1 hour.

Lock (Padlock)

Locks the control panel to prevent accidental changes during live events. When locked, all sliders, buttons, and configuration options are disabled. Click the padlock icon again to unlock.

API Reference

VVD exposes three network APIs for external control and automation. All APIs are available whenever VVD is running.

HTTP API (Port 8088)

RESTful API with JSON request and response format. CORS is enabled for browser-based integrations. Two API versions are available:

• V1 (Legacy): Original API endpoints maintained for backward compatibility.
• V2 (Full): Complete API with access to all VVD features including channel state, trigger control, mute/unmute, scene management, and configuration queries.

TCP API (Port 8089)

Persistent connection-based API using plain text commands. Ideal for applications that need to maintain a continuous link to VVD for real-time bidirectional communication. Supports event subscriptions for state change notifications.

UDP API (Port 8090)

Fire-and-forget, low-latency command interface. Best for time-critical trigger firing where connection establishment overhead is unacceptable. No acknowledgement is returned.

Common Operations

• Trigger a channel: Force-fire all triggers for a specific channel.
• Get channel status: Query which channel is currently active, audio levels, and detection state.
• Mute/Unmute: Temporarily disable or re-enable specific channels.
• Scene switching: Load a different scene via API command.
• Power on/off: Start or stop automatic switching.

Color Convention

VVD supports two colour conventions for tally and status indicators. Choose the convention that matches your industry:

Broadcast (TV / Radio)

RED = On Air, GREEN = Standby. This is the universal broadcast standard used in television studios and radio stations worldwide. A red tally light means the camera or microphone is live.

Live Events (Corporate AV)

GREEN = Active, RED = Off. Common in corporate AV, conference, and live event environments where green intuitively means “go” and red means “stop.”

System Requirements

• Operating System: Windows 10 or Windows 11 (64-bit)
• Runtime: .NET 8.0 Runtime
• CPU: Multi-core recommended; 4+ cores for 40+ channels
• GPU: Optional — NVIDIA, AMD, or Intel GPU for DirectML acceleration of AI voice detection
• RAM: 4 GB or more recommended
• Audio: At least one supported audio source (WASAPI device, ASIO driver, NDI source, or compatible console/DSP)