@bocoup/windows-sapi-tts-engine-for-automation

AT Driver

A WebSocket server which exposes the speech being vocalized by various screen readers running locally on a Microsoft Windows system.

Requirements

• Microsoft Windows
• Node.js

Installation

1. Install the project by executing the following command:

   npm install -g @bocoup/windows-sapi-tts-engine-for-automation
   

   If prompted for system administration permission, grant permission.

2. Start the server by executing the following command in a terminal:

   at-driver
   

   The process will write a message to the standard error stream when the WebSocket server is listening for connections. The --help flag will cause the command to output advanced usage instructions (e.g. at-driver --help).

3. Configure any screen reader to use the text-to-speech voice named "Bocoup Automation Voice."

4. Use any WebSocket client to connect to the server specifying v1.aria-at.bocoup.com as the sub-protocol. The protocol is described below. (The server will print protocol messages to its standard error stream for diagnostic purposes only. Neither the format nor the availability of this output is guaranteed, making it inappropriate for external use.)

Terminology

• message - a JSON-formatted string that describes some occurrence of interest, emitted at the moment it occurred; the message should be a JSON object value with two string properties: type and data
• message type - one of "lifecycle", "speech", or "error"
  • "lifecycle" - signifies that the message data is an expected lifecycle of the automation voice (e.g. initialization and destruction)
  • "speech" - signifies that the message data is text which a screen reader has requested the operating system annunciate
  • "error" - signifies that an exceptional circumstances has occurred
• message data - information which refines the meaning of the message type

Protocol

This project uses an application-level protocol named v1.aria-at.bocoup.com to communicate with clients via a WebSocket connection. All messages are encoded as JSON text.

// Clients may send Command messages to the server at any time. The server will
// respond to every Command it receives with a corresponding Response whose
// `id` value matches that of the Command which initiated it. The client may
// use any numeric value to uniquely identify the Command and to correlate the
// eventual Response.
interface PressKeyCommand {
  type: 'command';
  id: number;
  name: 'pressKey';
  params: [string];
}

interface ReleaseKeyCommand {
  type: 'command';
  id: number;
  name: 'releaseKey';
  params: [string];
}

interface SuccessResponse {
  type: 'response';
  id: number;
  result: any;
}

interface ErrorResponse {
  type: 'response';
  id: number;
  error: string;
  message: string;
}

interface SpeechEvent {
  type: 'event';
  name: 'speech';
  data: string;
}

interface LifecycleEvent {
  type: 'event';
  name: 'lifecycle';
  data: string;
}

interface InternalErrorEvent {
  type: 'event';
  name: 'internalError';
  data: string;
}

Architecture

This tool is comprised of two main components: a text-to-speech voice and a WebSocket server.

Text-to-speech voice

The text-to-speech voice is written in C++ and integrates with the Microsoft Speech API (SAPI). Because it interfaces with the Windows operating system (that is: "below" the screen reader in the metaphorical software stack), it can observe speech from many screen readers without coupling to any particular screen reader.

The voice has two responsibilities. First, it emits the observed speech data and related events to a Windows named pipe. This allows the second component to present a robust public interface for programmatic consumption of the data. (The named pipe is an implementation detail. Neither its content nor its presence is guaranteed, making it inappropriate for external use.)

Second, the voice annunciates speech data. It does this by forwarding speech data to the system's default text-to-speech voice. This ensures that a system configured to use the voice remains accessible to screen reader users.

WebSocket server

The WebSocket server is written in Node.js and allows an arbitrary number of clients to observe events on a standard interface. It has been designed as an approximation of an interface that may be exposed directly by screen readers in the future.

License

Licensed under the terms of the MIT Expat License; the complete text is available in the LICENSE file.

Copyright for portions of AT Driver are held by Microsoft as part of the "Sample Text-to-Speech Engine and MakeVoice" project. All other copyright for AT Driver are held by Bocoup.