I've just downloaded the new Piper TTS app onto my iPhone 17. I cannot find how to activate it within the settings. It doesn't appear anywhere in the list of available voices. Anyone got any idea how to make use of this app.
I'm waiting for Piper to get less nerdish and broken, and I'll it a go.
Have you checked out TGSpeechBox? It's multiplatform and open-source: https://github.com/tgeczy/TGSpeechBox/releases
There's Testflight and Ap Store versions. Tomi's been doing a fantastic job.
You just need to download voices from within the Piper app and they will appear everywhere. App by itself doesn't have a single voice if you just install it.
To be honest it is fairly fine product and people are overreacting in several ways:
- like in the comment above, claiming that it is nerdy (not sure why, because you need to download voices, and the app that he suggests that needs to be compiled and side loaded to run isn't nerdy at all...)
- trying it on the low-end hardware that can't handle it (yes you likely shouldn't try below iPhone 15 Pro)
- using highest quality voices from app instead of low quality (I agree that here you can blame the author for including them at all as they likely need at least latest iPad Pro to run properly on iOS) - just to note that what they call "high" quality is more like professional podcasts quality which no one expects.
Lost unsaved post and furious now; can’t rewrite. Stop complaining and reach out to the dev if you have really severe issues, just like I did. He did fix some stuff. And Speech Central dev gives constructive feedback and does his best to help us instead of pushing Acapela to keep their voices exclusive to his app and then charging us for the voices. He even promotes a totally free, humanlike and fairly responsive set of voices like these instead of looking for ways to exploit our needs.
@speech Central, I'm using an iphone 15 Pro with iOS 26 beta. The Piper voices I've tried are quite laggy and speak at a snail's pace. Not my problem if something doesn't work correctly with my configuration.
Also, tgSpeechBox has been on the standare app store for a while, so theres no sideload needed.
Plus I hate natural sounding voices for screen reading, but also wanted to explore and see how this project was doing. Would be cool to have something more natural to get that latest book chapter read, or whatever.
According to another thread on this very forum, Piper is still rather buggy. It also could depend on which voices are installed, ios and iphone make, etc. For those who get along with it, fantastic. I wish nothing but the best for this project going forward. I just found it unusable with my current phone and iPad.
However every AI voice will be a very bad VoiceOver voice for most people and if you tried it that way I fully understand you. Just you should expect it not to work well from the very start.
The speed problem is likely something mentioned on AppleVis alread - currently speed commands are reversed and if you speed up it will slow down.
However for general narrator use case (again not screen reader) those voices are very good. I notice just one problem and that is that it works bad with apostrophe which is hardly acceptable in English as it appears to frequently and may be confusing.
The text-to-speech (TTS) system requires significant improvements in grammar structures and pronunciation accuracy.
For instance, instead of saying “Jon’s car,” the TTS will incorrectly pronounce it as “John’ S car.” This error can be easily corrected by ensuring proper apostrophe usage.
Similarly, instead of saying “$1300,” the TTS will incorrectly pronounce it as “dollar 1300.” This error can be corrected by ensuring proper currency pronunciation.
Another common complaint is that TTS systems, except for the eloquence TTS, tend to pronounce numbers in pairs rather than full numbers. For example, instead of saying “1300,” they will say “thirteen hundred.” This is because we typically do not say “1300” in English; instead, we say “thirteen hundred.”
Additionally, TTS systems on Speech Market do not consistently pronounce dates as they are formatted. For instance, “11 September” will always be pronounced as “September 11” in the American TTS, while “September 11” will always be pronounced as “11 September” in the British English TTS.
The only TTS system on Apple OSES that consistently pronounces dates in the correct format is Alex. I will provide examples of both date formats for users to try.
**European Format:** 1st January, 2026
**American Format:** January 1st, 2026
In general synthesizer provider should fix it. As it was mostly Apple, you are right on that historically. However as now there are 3rd party synthesizers, it is not always Apple.
Here we may get too technical.
The Piper app has three components - piper synthesizer, eSpeak NG phoneme parsers, app interface.
When we talk about all of those problems, I think they all appear to originate from eSpeak NG phoneme parsers, which are limited, but they are the only code that is available on iOS to be used for this purpose by developers.
In that regard, it is a bit strange that Apple did not provide its own phoneme parser to synthesizer extensions, as they likely exist. That would make making of 3rd party synthesizer much more trivial than it is now.
Either way, I don't think that you can benchmark those problems on the same scale - as I said I do consider parsing of apostrophe critical, but I don't and can't consider that for currency, especially for use case of narration. Most of the text don't have currency at all and if they do current solution makes it a bit odd, but not something completely confusing. If you compare that to strict robotic reading that may have this feature done right, I really can't see how that would be acceptable trade-off in favor of robotic reading for most users. Again everyone may have his opinion and maybe in every text that he reads currency appears and all of that is perfectly legitimate, just I can't see that as dominant user behavior.
Has anyone else noticed that the words sound clipped off, especially as you increase the rate? If it wasn't for this and the issue with apostrophes, I would totally use some of these voices for background reading apps like Speech Central.
The eSpeak-NG app and engine work fine though, so can’t the developer of Piper TTS just replace whatever breaks Piper with whatever makes eSpeak work well?
Comments
Check this thread
https://applevis.com/apps/ios/utilities/piper-neural-tts
Quite a few people have had problems with this TTS engine. 🤷
i'm waiting
I'm waiting for Piper to get less nerdish and broken, and I'll it a go.
Have you checked out TGSpeechBox? It's multiplatform and open-source:
https://github.com/tgeczy/TGSpeechBox/releases
There's Testflight and Ap Store versions. Tomi's been doing a fantastic job.
Simple
You just need to download voices from within the Piper app and they will appear everywhere. App by itself doesn't have a single voice if you just install it.
To be honest it is fairly fine product and people are overreacting in several ways:
- like in the comment above, claiming that it is nerdy (not sure why, because you need to download voices, and the app that he suggests that needs to be compiled and side loaded to run isn't nerdy at all...)
- trying it on the low-end hardware that can't handle it (yes you likely shouldn't try below iPhone 15 Pro)
- using highest quality voices from app instead of low quality (I agree that here you can blame the author for including them at all as they likely need at least latest iPad Pro to run properly on iOS) - just to note that what they call "high" quality is more like professional podcasts quality which no one expects.
Stop Complaining
Lost unsaved post and furious now; can’t rewrite. Stop complaining and reach out to the dev if you have really severe issues, just like I did. He did fix some stuff. And Speech Central dev gives constructive feedback and does his best to help us instead of pushing Acapela to keep their voices exclusive to his app and then charging us for the voices. He even promotes a totally free, humanlike and fairly responsive set of voices like these instead of looking for ways to exploit our needs.
My experience may be different than yours
@speech Central, I'm using an iphone 15 Pro with iOS 26 beta. The Piper voices I've tried are quite laggy and speak at a snail's pace. Not my problem if something doesn't work correctly with my configuration.
Also, tgSpeechBox has been on the standare app store for a while, so theres no sideload needed.
Plus I hate natural sounding voices for screen reading, but also wanted to explore and see how this project was doing. Would be cool to have something more natural to get that latest book chapter read, or whatever.
According to another thread on this very forum, Piper is still rather buggy. It also could depend on which voices are installed, ios and iphone make, etc. For those who get along with it, fantastic. I wish nothing but the best for this project going forward. I just found it unusable with my current phone and iPad.
Re:
The lag is very good for most use cases.
However every AI voice will be a very bad VoiceOver voice for most people and if you tried it that way I fully understand you. Just you should expect it not to work well from the very start.
The speed problem is likely something mentioned on AppleVis alread - currently speed commands are reversed and if you speed up it will slow down.
However for general narrator use case (again not screen reader) those voices are very good. I notice just one problem and that is that it works bad with apostrophe which is hardly acceptable in English as it appears to frequently and may be confusing.
This voice needs a lot of improvement.
The text-to-speech (TTS) system requires significant improvements in grammar structures and pronunciation accuracy.
For instance, instead of saying “Jon’s car,” the TTS will incorrectly pronounce it as “John’ S car.” This error can be easily corrected by ensuring proper apostrophe usage.
Similarly, instead of saying “$1300,” the TTS will incorrectly pronounce it as “dollar 1300.” This error can be corrected by ensuring proper currency pronunciation.
Another common complaint is that TTS systems, except for the eloquence TTS, tend to pronounce numbers in pairs rather than full numbers. For example, instead of saying “1300,” they will say “thirteen hundred.” This is because we typically do not say “1300” in English; instead, we say “thirteen hundred.”
Additionally, TTS systems on Speech Market do not consistently pronounce dates as they are formatted. For instance, “11 September” will always be pronounced as “September 11” in the American TTS, while “September 11” will always be pronounced as “11 September” in the British English TTS.
The only TTS system on Apple OSES that consistently pronounces dates in the correct format is Alex. I will provide examples of both date formats for users to try.
**European Format:** 1st January, 2026
**American Format:** January 1st, 2026
Apple should fix that; not the devs.
roman was right to some extent
In general synthesizer provider should fix it. As it was mostly Apple, you are right on that historically. However as now there are 3rd party synthesizers, it is not always Apple.
Here we may get too technical.
The Piper app has three components - piper synthesizer, eSpeak NG phoneme parsers, app interface.
When we talk about all of those problems, I think they all appear to originate from eSpeak NG phoneme parsers, which are limited, but they are the only code that is available on iOS to be used for this purpose by developers.
In that regard, it is a bit strange that Apple did not provide its own phoneme parser to synthesizer extensions, as they likely exist. That would make making of 3rd party synthesizer much more trivial than it is now.
Either way, I don't think that you can benchmark those problems on the same scale - as I said I do consider parsing of apostrophe critical, but I don't and can't consider that for currency, especially for use case of narration. Most of the text don't have currency at all and if they do current solution makes it a bit odd, but not something completely confusing. If you compare that to strict robotic reading that may have this feature done right, I really can't see how that would be acceptable trade-off in favor of robotic reading for most users. Again everyone may have his opinion and maybe in every text that he reads currency appears and all of that is perfectly legitimate, just I can't see that as dominant user behavior.
words Clipping
Has anyone else noticed that the words sound clipped off, especially as you increase the rate? If it wasn't for this and the issue with apostrophes, I would totally use some of these voices for background reading apps like Speech Central.
@Speech Central
The eSpeak-NG app and engine work fine though, so can’t the developer of Piper TTS just replace whatever breaks Piper with whatever makes eSpeak work well?