Example for integrating a third-party speech synthesizer

[1ec5]

Plenty of text-to-speech engines besides Amazon Polly support SSML. Here’s a partial list. RouteVoiceController is open, so a developer write an adapter for a text-to-speech engine of their choice by subclassing it and overriding a few methods, using PollyVoiceController as a model. But an example would make it much easier for someone to do that.

/cc @bsudekum @ericdeveloper @captainbarbosa

[bsudekum]

Once https://github.com/mapbox/mapbox-navigation-ios/pull/617 is live, do you think there is much of a use case? Off the cuff, I think this would be a rather big lift for us.

[1ec5]

Once mapbox/mapbox-navigation-ios#617 is live, do you think there is much of a use case?

Yes. There will remain some (hopefully niche) reasons to use something other than Polly via the Mapbox Speech API:

• An application may target a market that speaks a language that Polly does not and that AVSpeechSynthesizer sounds horrible on.
• Neither Polly nor AVSpeechSynthesizer supports Catalan or Vietnamese.
• There’s a cottage industry of TTS engines focused on Chinese for applications running within the PRC. We haven’t focused on a CN endpoint thus far.
• An application may already integrate another speech library for a non-navigation-related feature or for the library’s speech recognition functionality.

Whatever the case, we don’t have to get too fancy with this example. Just take MapboxVoiceController and replace all the code specific to the Mapbox Speech API with a comment explaining what TTS functionality to hook into. It can be up to the developer to understand the third-party library; we don’t have to advertise a particular library.

[1ec5]

a developer write an adapter for a text-to-speech engine of their choice by subclassing it and overriding a few methods, using PollyVoiceController as a model.

As of mapbox/mapbox-navigation-ios#617, the developer would subclass RouteVoiceController, overriding didPassSpokenInstructionPoint(_:) (to prefetch audio from the network, if desired) and speak(_:) (to play the audio). To keep things simple, the example would probably assume the audio files have been bundled with the application, so the subclass wouldn’t need to override didPassSpokenInstructionPoint(_:); all it would do is override speak(_:) to create an AVAudioPlayer from the file and play it.

[1ec5]

The override would look something like this:

https://github.com/mapbox/mapbox-navigation-ios/blob/c026c6fe76c34f85c80e50324e6ff4e8f422f256/MapboxNavigation/MapboxSpeechController.swift#L173-L198

I suppose we could also factor out an AudioPlayerVoiceController or something to make this task easier, since methods like duckAudio() are private.