Reference Applications

VEE Audio ships with a set of ready-to-use reference applications built on the framework. They can be deployed individually or composed into a single firmware image using the all-in-one application. Each application serves both as an example and as a starting point for custom development.

Reference Voice Assistants

Voice Assistant applications implement the VoiceAssistant interface and act as the voice interface between the user and all other installed applications. Three reference implementations are provided:

ChatGPT

Backed by the OpenAI Realtime API. Provides conversational responses, general knowledge queries, and function dispatching to installed applications.

Voice Live

Backed by the Azure AI Foundry Voice Live API. An alternative Voice Assistant with enterprise-grade speech synthesis and recognition, interchangeable with ChatGPT at configuration time.

Qwen

Backed by the Alibaba Cloud Model Studio (DashScope) compatible-mode realtime API. Designed for deployments in regions or contexts where Alibaba Cloud AI services are preferred.

Audio Applications

Audio applications implement audio features that the Voice Assistant can control through Function Calling. Each application registers its Functions with the Kernel at startup; the active Voice Assistant picks them up automatically.

Web Radio

Streams live audio from internet radio services. Exposes functions to start a named radio station, list available stations, and stop playback. Uses the Audio Focus system so that radio playback yields when a higher-priority audio event occurs.

Voice Note

Records the user’s voice and writes down what they dictate. The dictation is transcribed through the active Voice Assistant and the resulting text is sent to a companion application. Useful as a hands-free note-taking tool.

Web Search

Performs a web search from a voice query and reads back a summary of the results. It embeds no AI client of its own: it submits the query to the active Voice Assistant through the AI Query API, whose AI backend searches the web and returns a summarized answer, spoken back to the user.

Volume

Controls the system audio output level by voice command. Exposes functions to set an absolute volume level, increase or decrease the volume by a step, and mute or unmute the output. Demonstrates the use of the volume control of the AudioService.

Alarm

Sets and triggers a timed audio alarm. The user sets a delay by voice; after the delay expires, the alarm plays a repeating sound at maximum priority, preempting any other audio. Demonstrates focus-level event handling: a button press dismisses the alarm regardless of which application is currently active.

Composite Deployment

The all-in-one application is a convenience wrapper that bundles the Kernel together with all reference applications into a single deployable artifact. It auto-generates the required target configuration files and can be built and deployed with a single command. By default, ChatGPT is the active Voice Assistant; Voice Live or Qwen can be substituted at configuration time.

This composite mode is the recommended starting point for evaluation and prototyping. For production, applications are typically deployed as separate Feature packages, giving full control over which capabilities are included in the firmware image and which are deployed dynamically over the air.

Example Session

The following transcript illustrates a typical interaction with the all-in-one firmware, showing how the Voice Assistant, Audio Focus arbitration, and several reference applications cooperate within a single session.

[music streaming over Bluetooth A2DP]

User            > presses the Voice Assistant button
[Bluetooth music pauses: the Voice Assistant takes audio focus]
Voice Assistant > "Hi, how can I help you?"

User            > "What's the weather like in Nantes today?"
Voice Assistant > "Searching the web..."
Voice Assistant > "Today in Nantes, expect ..."

User            > "Listen to BBC"
Voice Assistant > [dispatches the request to the Web Radio application]
Web Radio       > [streams BBC]

This single session exercises Audio Focus arbitration (the Bluetooth stream yields to the Voice Assistant), the AI Query API (the Web Search application queries the AI backend), and Function Calling (the Web Radio application is started by voice).