2018-06-19 22:26:27

Hi all,
I heard that there were some new Siri voices released recently, which I wanted to try out. After playing with them for a bit, I found that they had some pronunciation issues which I wasn't happy with. Ultimately, I decided to go back to using the US Siri male, the same voice I've been using for 2 years.

When I did so, I found that I was not presented with an option to choose between the compact and enhanced variants. I got stuck with the enhanced one, which sounds awful to me, it always did. The other voices, such as Samantha, Tom, etc. still provide the option to switch, and it works just fine.

Is this a bug in the latest version of IOS, or is there something else I can do besides resetting my phone to get it back? If it is a bug, I'll be more than happy to submit a report, because this sucks. It's the only voice I can stand listening to for long periods of time. As I said, I'm using the latest version of IOS 11, and before today, had not changed the voice at all for nearly 2 years, so for all I know, this could be either a long-standing problem or a new feature, which I certainly hope it isn't. Similarly, I would be willing to bet the new Siri voices sound much better with their compact variants anyway.

The glass is neither half empty nor half full. It's just holding half the amount it can potentially hold.

2018-06-20 03:38:02 (edited by Slender 2018-06-20 03:38:21)

I don't think there are compact Siri voices anymore, unfortunately, or at least none that can be used by Voiceover. Oddly there is a compact Siri voice available, but Voiceover can't use it for some reason.

Oh no! Somebody released the h key! Everybody run and hide!

2018-06-20 04:49:33

That doesn't make any sense. Where did it go? I was just using it this morning! Clearly the voice has to still be stored somewhere on my phone.

The glass is neither half empty nor half full. It's just holding half the amount it can potentially hold.

2018-06-20 09:16:20

I redownloaded the Siri Female Voice US, but it didn't have any compact version.
Which are the new voices by the way?

2018-06-20 14:51:52

iOS handles voices in a strange way I think. It has a compact fallback that it uses in case the voice is unavailable, then it switches to an enhanced variant when it downloads the voice. What were you using the compact voice with?

Oh no! Somebody released the h key! Everybody run and hide!

2018-06-20 17:21:52

I was using the compact voice with Voiceover. What doesn't make sense to me is why the enhanced Siri voices sound great when actually used with Siri, but sound robotic and have tons of pronunciation issues when used with Voiceover. I'm writing to Apple Accessibility now to see if something can be done about this.

The glass is neither half empty nor half full. It's just holding half the amount it can potentially hold.

2018-06-20 21:20:31 (edited by musicalman 2018-06-20 21:24:48)

Turtlepower, are you talking about the new South African English Siri voices as being the new ones you wanted to try out? If so, they do have weird issues. I like them though, so once those get fixed I may use them.

I am actually the opposite as you, I really do hate all compact voices. They sound raspy/rubbery and metallic to me, and some of them dont' have much clarity. If I have to use a compact voice, Samantha will do. But I always use the enhanced US Siri male voice.

In IOS 10, a lot of things changed with Siri voices to increase their naturalness. The intonation on the IOS 10 version was a little robotic for me, but I still wanted to try it in Voiceover. Strangely you couldn't use those versions with Voiceover though. When you downloaded the enhanced voices, it would fallback to the IOS 9 versions which, while being a little more lively sounding to me, they were of lower quality and had annoying artifacts I didn't like. I used it because it was my favorite at the time. I can't remember if compact Siri voices were available in IOS 10, but if I had to guess, I'd bet the old ones were still available.

But at some point the compact Siri voices seemed to have been removed. This was definitely the case in IOS 11, since when I started trying betas, the compact Siri voices weren't there. I'm pretty sure their removal coincides with Apple taking charge of development and using a different engine; Nuance previously just used the Vocalizer engine for the Siri voices which is why they could so easily work with the NvDA Vocalizer driver. Why they'd wait for IOS 11 to fully switch to the new engine is beyond me, but maybe it wasn't stable enough for Voiceover or something. That could well be, since when they first became available to Voiceover they indeed had really bad bugs. In any case, since siri voices use a different engine now, I doubt compact voices will exist anymore. Given Siri's purpose, I can understand why, though the way they quietly handled it is a little odd and unfortunately it directly affects people like you which isn't right. I'm not sure if new versions of IOS fall back on a compact voice if the enhanced ones aren't available, or if they just use Samantha or something now. But, if you were still using the old compact voices, or the old enhanced voices from IOS 9, those old versions will still work even on IOS 11 until you switch voices. At least that was the case for me, and it would make a lot of sense as to why you are only noticing this now. If you used the compact version for two years without a switch, nothing was forcing you not to use it. Once you switch though, the old versions that you were using are no longer available. Whether they are still on the phone is anyone's guess. They may be there, but then again they may not, you never know with Apple.

What pronunciation and robotic issues do you have with the enhanced US male? You said that it sounds fine in Siri but in Voiceover it is not to your liking? Can you detail this a little more? Can you specify the differences you hear, or provide a recording of how it sounds in Voiceover? There shouldn't be any differences, though I haven't done any real hard testing to confirm this. ON mine though, they are very similar and that is why I use them.

Make more of less, that way you won't make less of more!
If you like what you're reading, please give a thumbs-up.

2018-06-20 21:52:20 (edited by Slender 2018-06-20 21:57:40)

Interestingly, in the early iOS 11 betas, all old Siri voices were available to be downloaded, including Aaron, Nicky and even Martha. Then they disappeared and the only old voice remaining now is Nicky, and that seems to be the 22050Hz version, not the 44100Hz version from iOS 10 Siri. @7, about the Vocalizer engine, it seems that they might still use it to an extent in the current version of Siri, but of course not a Nuance provided voice. Siri accepts Vocalizer speech tags.

Oh no! Somebody released the h key! Everybody run and hide!

2018-06-20 22:08:29

Oh? I didn't know Vocalizer tags were accepted in Voiceover, or that Siri voices supported them. I don't use them. As an aside, I do think it would be interesting to see all the other Vocalizer voices get that kind of upgrade. Granted they'd be quite a bit larger, and compact voices would still have to be a thing in those cases since those voices are still used a lot in imbedded systems I think where enhanced versions would be too much, but I think we're at a place now where the benefits of higher than 22K voices outweigh the costs for the most part, and besides with the new machine learning they're doing, the unit selection seems to be a little smarter. Speaking of which, I read this interesting article explaining the technicalities of Siri voice evolution with a few audio samples to illustrate the changes. Nothing mentioned about compact voices, but the article is pretty interesting as a fan of the enhanced voices.

While looking for random stuff like that, I did find hints that IOS 12 will bring a new Siri voice. Any info on that?

Make more of less, that way you won't make less of more!
If you like what you're reading, please give a thumbs-up.

2018-06-20 22:19:37 (edited by Slender 2018-06-20 22:23:18)

Yeah, Voiceover accepts them just like Macintalk tags, however they're much more difficult to write since you have to send an escape character first. And as for a new Siri voice, I can't find any new voices in the current developer beta. Though this is relatively early on in the beta cycle, so a new voice may very well pop up later, though I doubt it since new voices weren't demonstrated even at WWDC18.

Oh no! Somebody released the h key! Everybody run and hide!

2018-06-21 07:39:58

At Musicalman, yes, I was referring to the South African voices. The male voice in particular has a very pleasant accent, and, as I am rather partial to that accent in general, it motivated me to see how they stacked up. As you said, they could definitely be improved on, which I hope is the case. I probably wouldn't be complaining otherwise.

I'll see what I can do about a sample. One of the things I notice though, is that the US Siri male tends to stress punctuation in the wrong places. For example, if you're in settings, and you want to go into Voiceover's options, it will say Voiceover, on? Note the punctuation, as to me, there is not only an inappropriate pause between Voiceover and on, but why is it a question? The compact voice did not do this. It seems nitpicky, I know, but it just irks me.

He also sounds clearer when actually used with Siri, although this might be something an actual sample would be better at clarifying. I usually use the British Siri male voice with Siri, so I haven't actually done extensive testing, but from what I played with yesterday, it did seem to be the case that the enhanced voice, when used with Voiceover, had a muddier quality that I didn't like, and again, the inflection seemed off in certain places, although nothing I can exactly put my finger on at the moment. If it would help, I would be more than happy to upload a sample.

The glass is neither half empty nor half full. It's just holding half the amount it can potentially hold.

2018-06-21 08:44:40

TurtlePower is correct. i've also seen this weard issue that, if using iphone seven or newer, VO refuses to use the default brittish female voice, but rather uses the enhanced voice. wich sounds hooorible. it does not want to use the compact versions of voices for some languages.

There's a place for me in this universe.

2018-06-21 11:14:42

around IOS11 beta 4 or so, all IOS 10 siri voices were removed, along with the default. at least, as others have said, they may still be on your phone, but you won't be able to use them. there is an enhanced and a compact of everyone except all the siri voices, and alex. if you look at Nicki, for instance, even her compact version was removed. since, I'll assume you didn't touch your voice settings throughout IOS10, you weren't affected. unless you have an iCloud backup before the change, though, I don't see any ways for you to get back. was personally never a fan of compact voices here either. if I must, I'll take Samantha, but well. another thing. US and UK siri male voices can't handle fast rates at all, and while they kind of work, as others have also pointed out, they pause for no reason, insert question marks where they shouldn't, and quite a number of things. been using susan for quite a while over here, partly because no one uses her, got bored of Samantha, and Kate doesn't exactly have a nice tone. South african siri is pretty nice, (not a fan of male voices in general) and she does seem to be more responsive than the US siri female at this point.

2018-06-21 12:59:21

I particularly like the new South African female voice. She tends to speak in an RP accent all the time and you can hardly make any difference from British English, except for the pronunciation of the letter "R" when read in isolation.
But all in all, a great job. Well done Apple!
I may well use it as a primary voice.

2018-06-21 15:32:07 (edited by Slender 2018-06-21 15:35:23)

Turtlepower17, yeah, question intonation when there doesn't need to be any is actually somewhat common with Siri voices. Apple seems to be trying to get rid of the question words as they pop up, recently the words "Braille" and "words" when spoken on there own are not rising anymore. South African Siri used to have an issue with z as well, saying "zed? Zulu?" when it was encountered, but it's partially fixed in betas now. Also the phrase "English U.S." was fixed, though there was an interesting bug at one point where Daniel would say "English U, slnc 200, k", most likely because the speech manager was sending the wrong commands to the wrong synthesizer.

Oh no! Somebody released the h key! Everybody run and hide!

2018-06-21 15:38:39

Still waiting for the day when "natural" voices start inflecting on exclamation marks xD.

2018-06-21 21:14:58

Haha Munawar, indeed. Lyrebird and Google will probably be the firfst to do it, though something is giving me the impression that Google will be the first to offer that functionality in a commercial product like a general purpose TTS voice. Lyrebird doesn't seem to be geared at general purpose usage.
At Turtlepower, I just tested the US male and in your Voiceover On example, it didn't sound like a question but rather an indefinite period, like he knew he was supposed to end the phrase but wasn't sure if it was a full sentence or just part of one, a semicolon perhaps big_smile. But yeah, Siri voices do have weird intonation at times. I take it as part of the territory with these enhanced voices, but I can understand how it would be annoying if you're not used to it, and especially if you actually prefer compact.

Make more of less, that way you won't make less of more!
If you like what you're reading, please give a thumbs-up.

2018-06-22 01:57:35

The new Siri voices are using a different speech engine? I wonder if it's Macintalk? What are these Vocalizer tags? I'd like to learn them.

Grab my Adventure at C: stages Right here.

2018-06-22 04:22:13

Well, I'll see what, if anything, Apple has to say about the matter, but I'm beginning to see that I'm fighting a losing battle here, unfortunately. For now, Tom is the least hideous of the bunch, so I'm trying to take it in stride.

The glass is neither half empty nor half full. It's just holding half the amount it can potentially hold.

2018-06-22 11:24:49

Chris, I may have been mistaken about the engine being totally different. It's based on Vocalizer, all-be-it customized with the deep learning algorithms that are used for the Siri voices, and working at higher sampling rates than you get with standard Vocalizer. These changes are exclusive to Siri voice land, so that's probably why they can't be used anywhere else right now. AS for the tags, you'll have to ask Slender, or other people.

Make more of less, that way you won't make less of more!
If you like what you're reading, please give a thumbs-up.

2018-06-22 14:08:07

Yeah, the tags are a bit like SAPI 4, \attribute=value\, but they're more complicated to write because you have to send an escape character, character 27, 0x1b, . Pitch, rate and volume are written as something like \pitch=10\, or \volume=10\. Vocalizer also allows you to change the voice being used on the fly with \voice=name\, but I've never gotten that command to work on iOS, possibly because Apple's implementation may identify the voices differently.

Oh no! Somebody released the h key! Everybody run and hide!

2018-06-25 12:01:27

is there a way to download the sample of all the voices as mp3 or any other format?

He picked up the wrench and broke the guy’s wrist with it, one, and then the other wrist, two, and turned back and did the same to the guy who had held the hammer, three, four. The two men were somebody’s weapons, consciously deployed, and no soldier left an enemy’s abandoned ordnance on the field in working order.