A while ago I added the possibility to configure Seddit (my text-to-speech-focused hands-free Reddit client for macOS and iOS) with multiple voices so that each user’s content could be read in a different voice. Of course, iOS and macOS come with voices that speak a huge variety of different languages, so you could theoretically select, say, a Japanese voice, a French voice, and three English voices, and download Reddit posts and comments in all of those languages. However, until now, Seddit would randomly assign a voice to each user, without regard for the language that user had written in, so if you did that, you could end up with English posts pronounced as if they were French, Japanese character names read out in English, and so on.

In the latest version, if you select voices that speak multiple languages in the Voices tab of the Settings screen, when Seddit encounters a post or comment by a user it hasn’t chosen a voice for yet, it will detect which of those languages the post or comment is probably in, and choose a voice that knows how to pronounce that language.

Of course, this isn’t perfect — it still always uses the same voice for each user, so if a user sometimes posts in French, and sometimes in English, or if they write in multiple languages within a single post because it’s a language-learning subreddit, then some of that is going to be spoken using an inappropriate voice. Also, if someone only writes in English but the first comment that Seddit encounters of theirs is an image meme and the text ‘c’est la vie!’ Seddit might determine that the user speaks French, and then hilariously mispronounce the rest of their posts. Note, if there is not enough text in the user’s first post for Seddit to even guess the language, it will not definitively choose a voice for that user until it encounters another post by them. I have yet to find either of these situations in practice, even while looking for them, so I hope it’s a rare issue.

Nonetheless, all of these situations are better than Seddit just randomly picking a voice for each user, regardless of which language they happen to be writing in. You should try it out, especially if you want to listen to Reddit content in various languages!

I also redesigned the Settings screen on iOS and iPadOS so it’s fullscreen and has a close button in the top right, as per Apple’s human interface guidelines, instead of a ‘Done’ button taking up a lot of space at the bottom and making the tabs look weird.

Note, while writing this post, I tested the regular ‘Start Speaking’ menu command on macOS and the ‘Speak’ command on iOS and found that it will sometimes switch to appropriate voices if I select multilingual text, even if my System Speech Language est réglé sur えい語。 Okay, it doesn’t work well for the French/English parts of that sentence. Maybe it’s only good with switching between languages if I switch scripts, e.g. בַּרְוָזָן утконос カモノハシ. Yep, that works, although if I select any other text along with πλατύπους, it’ll read it as ‘Greek small letter pi’ etc. I guess Greek letters are used too often in English for the speech engine to assume we actually switched to Greek. There were certainly plenty of Greek letters in the Princeton Companion to Mathematics.

Anyhow, I’m thinking I could improve Seddit further by giving each user a voice in each language you’ve selected voices for, and detecting the language for each post/comment, or for each sentence. Though macOS doesn’t do that unless you switch scripts… when I tried adding ‘J’imagine qu’il choisit une nouvelle langue pour chaque phrase.’ as a separate sentence and selected it along with a few English sentences, it read the whole thing in a French voice.

On the subject of interesting text-to-speech behaviour, and interesting behaviour in general, remember my half-written Lola parody about Pitch Synchronous Overlap and Add? Well, the lovely Joey Marianer had an appointment in town a while ago, and sneakily recorded the song in a parking building as a surprise, because I’m usually home so there’s little chance to record things at home without my hearing. I was duly surprised and delighted. Even the disclaimer about the missing bridge sounds like it scans as a bridge! Now you can also be surprised, delighted, and probably confused as to why this half-baked song was considered worth singing.