Selectable audio input device

[compulim]

Fixes #2481. Fixes #2987. Fixes #2886. Fixed #2668.

Changelog Entry

Added

• Resolves #2481. Support selecting different audio input device for Cognitive Services Speech Services, by @compulim in PR #3079

Fixed

• Fixes #2668. Disable Web Audio on insecure connections, by @compulim in PR #3079

Changed

• Resolved #2886 and #2987, converged all references of microsoft-cognitiveservices-speech-sdk to reduce footprint, by @compulim in PR #3079

Samples

• Resolves #2481, added selectable audio input device sample, by @compulim in PR #3079

Description

Support selectable audio input device. From user report, even AirPod is connected, on iOS, it is not selected as default. JavaScript is required to select the device and make it default.

Also, today, we have 3 references of microsoft-cognitiveservices-speech-sdk, which take up significantly amount of size in our bundle (~300 KB each). We are going to converge them into a single copy inside the bundle.

We will temporarily bump to a preview build of web-speech-cognitive-services. When we do production release, we will build a production version of web-speech-cognitive-services and bump the version in Web Chat.

Specific Changes

• Update webpack.config.js to reference to the same microsoft-cognitiveservices-speech-sdk package to reduce footprint by ~600 KB
• Add audioContext option to createCognitiveServicesSpeechServicesPonyfillFactory()
  • This is for speech synthesis and supporting our test infrastructure, we will send in a mock audioContext object
• Add audioInputDeviceId option to createCognitiveServicesSpeechServicesPonyfillFactory()
  • If no audioConfig is set, we will use audioInputDeviceId to select the correct audio input device
• Add samples to show how to select audio input device, and a comprehensive sample to try out this feature
• Update HTML test infrastructure to enable testing under IE11
  • Some packages are not transpiled to ES5. We added a postinstall script to transpile them after npm run bootstrap or lerna add. We prefer to use postinstall because build time is substantial and to reduce the tax on our build
  • These packages were not transpiled to ES5:
    • @babel/plugin-proposal-async-generation-functions
      • This package is mainly used under Node.js, hardly in a browser, and rarely on IE11
    • event-iterator
      • This package is using async iterator, which is not natively supported on IE11
    • microsoft-cognitiveservices-speech-sdk
• Bump react-say to not breaking IE11

External changes

• https://github.com/compulim/react-say/compare/v2.0.0..c5d6304
• https://github.com/compulim/web-speech-cognitive-services/compare/v6.3.0..da7babf

---

• Testing Added

[corinagum]

To discuss:

• number of packages published for this project that aren't peer reviewed
• how complicated our single-page samples have gotten. They need to be simpler and split into re-used modules

[compulim]

Tests for Direct Line Speech added. Test is currently failing due to reliability, will fix it.