Last active
October 12, 2022 13:47
-
-
Save ryanlintott/e1284682216d4e4a90fa474a676fb453 to your computer and use it in GitHub Desktop.
Extension to AVSpeechSynthesizer that will speak IPA (International Phonetic Alphabet) strings
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| import Foundation | |
| import AVFoundation | |
| @available(iOS 10.0, *) | |
| extension AVSpeechSynthesizer { | |
| func speakIPA(_ ipaString: String, voiceIdentifier: String, willSpeak: ((String) -> Void)? = nil) { | |
| //Set the audio session to playback to ignore mute switch on device | |
| do { | |
| try AVAudioSession.sharedInstance().setCategory(AVAudioSession.Category.playback, options: [.interruptSpokenAudioAndMixWithOthers, .duckOthers]) | |
| } catch { | |
| print("Error: \(error.localizedDescription)") | |
| } | |
| let mutableAttributedString = NSMutableAttributedString(string: ipaString) | |
| let range = NSString(string: ipaString).range(of: ipaString) | |
| let pronunciationKey = NSAttributedString.Key(rawValue: AVSpeechSynthesisIPANotationAttribute) | |
| mutableAttributedString.setAttributes([pronunciationKey: ipaString], range: range) | |
| let utterance = AVSpeechUtterance(attributedString: mutableAttributedString) | |
| let voice = AVSpeechSynthesisVoice(identifier: voiceIdentifier) | |
| utterance.voice = voice | |
| // Pausing first is safer and may prevent bugs | |
| self.pauseSpeaking(at: .immediate) | |
| self.stopSpeaking(at: .immediate) | |
| // Run some code just before speaking | |
| willSpeak?(utterance.speechString) | |
| print("speakIPA: \(ipaString) voice: \(voice?.identifier ?? "?")") | |
| self.speak(utterance) | |
| } | |
| } |
ill definitely check it out.
just soent 2 hours listening to alex say (latin alpha colon) or alternating pronounciation of the exact same text and ipa
ie ala as ahlah alay at random.
ive tried several IPA notations and the only thing it seems to obey is the stress marker? ‘.
Just dove into this all yesterday.
Very annoying we dont have control over phonmrmes.
Some bg, im trying to get tts to read Toki pona which is already in ipa ie it is read as the ipa pronunciation.
There is very little on this online aside from a few ild firum posts ending in “i filed a bug” and in apple fashion years later still there.
Im very frustrated right now.
Thanks for the info
…Sent from my iPhone
On Apr 21, 2022, at 9:42 AM, Ryan Lintott ***@***.***> wrote:
***@***.*** commented on this gist.
This code is a bit older and I found a few bugs since. Mainly if you have a different region set on your phone the IPA might sound very strange. The latest version that works better is included in this package > OEVoice
—
Reply to this email directly, view it on GitHub, or unsubscribe.
You are receiving this because you commented.
I guess my confusion is that if the IPA attribute is for specifying pronunciation,
than any voice that has those phonemes should be able to pronounce it.
The consonants are the dame as english, vowels are long, and j is IPA j aka english y.
Given that, idk why it fails so badly and acts randomly.
that was with your snippet and my own attenpts.
…Sent from my iPhone
On Apr 21, 2022, at 11:29 AM, Jimmy Hough Jr ***@***.***> wrote:
ill definitely check it out.
just soent 2 hours listening to alex say (latin alpha colon) or alternating pronounciation of the exact same text and ipa
ie ala as ahlah alay at random.
ive tried several IPA notations and the only thing it seems to obey is the stress marker? ‘.
Just dove into this all yesterday.
Very annoying we dont have control over phonmrmes.
Some bg, im trying to get tts to read Toki pona which is already in ipa ie it is read as the ipa pronunciation.
There is very little on this online aside from a few ild firum posts ending in “i filed a bug” and in apple fashion years later still there.
Im very frustrated right now.
Thanks for the info
Sent from my iPhone
>> On Apr 21, 2022, at 9:42 AM, Ryan Lintott ***@***.***> wrote:
>>
> ***@***.*** commented on this gist.
> This code is a bit older and I found a few bugs since. Mainly if you have a different region set on your phone the IPA might sound very strange. The latest version that works better is included in this package > OEVoice
>
> —
> Reply to this email directly, view it on GitHub, or unsubscribe.
> You are receiving this because you commented.
Apple's version of IPA is quite limited. Each voice is mostly limited to the sounds used in the language and accent it was created for. I started with the voice that worked best initially then used a bunch of substitutions to alter the base OE IPA pronunciations and fix various errors as I found them. Often combinations of symbols just wouldn't work so I would swap with similar sounds until I got something close enough. If you check out this file you'll see all the substitutions I used to fine tune the pronunciation. OEVoice.swift
Interesting. I'm looking at it now.
What im working with feels like it should only need 2 changes for an English voice.
let consonantCharset = CharacterSet(charactersIn: "jklmnpstw")
let vowelCharset = CharacterSet(charactersIn: "aeiou")
All consonants are like English, save j which is IPA j.
Vowels are ah eh ee oh oo
Old English is further from standard English by a lot I think. So Im wanting to find the simplest solution.
I thought adding ipa for every world was tedious but simple and proper.
I guess what im seeing is, instead of using the IPA sounds for symbols in the IPA string,
It treats IPA symbols like normal English letters., which is what it does with the non attributed text.
It looks like you are replacing standard IPA symbols with something the voice will apply with the ipa attribute?
That feels like it should be close to normal English but apparently due to apple IPA being special
I’m not sure what to do. Im not sure about the other open source tts stuff available its been a decade since I used them and a quick search showed they didnt seem to be very swifty.
It just occurred to me, and it may interest you that there used to be under additional developer tool downloads
At the dev portal, there were utilities for developing a custom tts voice.
I wonder if that still exists, and would give insight to how the existing voices work.
Why it cant do what it says on the tin and just apply IPA rules to ranges idk.
Sorry for being so wordy, I often work completely solo and this problem is pretty niche to find any guidance on.
I did notice changing /a/ to /ah/ made it long, but produced weird artifacts in akala adding a yuh sound.
Got to be an easier way!
… On Apr 21, 2022, at 12:33 PM, Ryan Lintott ***@***.***> wrote:
@ryanlintott commented on this gist.
Apple's version of IPA is quite limited. Each voice is mostly limited to the sounds used in the language and accent it was created for. I started with the voice that worked best initially then used a bunch of substitutions to alter the base OE IPA pronunciations and fix various errors as I found them. Often combinations of symbols just wouldn't work so I would swap with similar sounds until I got something close enough. If you check out this file you'll see all the substitutions I used to fine tune the pronunciation. OEVoice.swift <https://github.com/ryanlintott/OEVoice/blob/main/Sources/OEVoice/OEVoice.swift>
—
Reply to this email directly, view it on GitHub <https://gist.github.com/e1284682216d4e4a90fa474a676fb453#gistcomment-4140754>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAL5QCLCGZ5GDFJAKUUHRXTVGGGOHANCNFSM5T7LAM4A>.
You are receiving this because you commented.
The only way to ensure your pronunciations are correct would be to add IPA pronunciations to each word. You need to use IPA symbols (I find this helpful: https://ipa.typeit.org) and there will be some combinations that Apple might not pronounce correctly. If Toki Pona is purely phonetic then maybe you could do a substitution swapping letters for IPA symbols. If however it's more like English where letters may sound different depending on the context you'll probably need to translate each word as the rules would be too complex to code. Either way, once you have IPA pronunciations you think are accurate, then I would try testing with voices, finding one that is the closest fit, then testing with words and making any voice-specific adjustments to IPA characters (like I've done with Old English).
When picking IPA symbols I found this app very helpful as you could hear the different sounds:
https://apps.apple.com/app/id869642260
Also this web app was helpful for testing even though it's using non-Apple voices as it's pretty accurate.
http://ipa-reader.xyz
Man, kinda OT but the app I was using this in got lost and now I have to rewrite it.
I gotta keep all my code in a remote now!
I think I had gotten something close enough with a Mexican Spanish voice with just a few overrides for double ll sound.
I still haven’t gotten the energy to rewrite that app, yet.
I appreciate the pointers and will revisit them when I rewrite toki pwnage.
… On Oct 11, 2022, at 12:49 PM, Ryan Lintott ***@***.***> wrote:
@ryanlintott commented on this gist.
The only way to ensure your pronunciations are correct would be to add IPA pronunciations to each word. You need to use IPA symbols (I find this helpful: https://ipa.typeit.org <https://ipa.typeit.org/>) and there will be some combinations that Apple might not pronounce correctly. If Toki Pona is purely phonetic then maybe you could do a substitution swapping letters for IPA symbols. If however it's more like English where letters may sound different depending on the context you'll probably need to translate each word as the rules would be too complex to code. Either way, once you have IPA pronunciations you think are accurate, then I would try testing with voices, finding one that is the closest fit, then testing with words and making any voice-specific adjustments to IPA characters (like I've done with Old English).
When picking IPA symbols I found this app very helpful as you could hear the different sounds:
https://apps.apple.com/app/id869642260
Also this web app was helpful for testing even though it's using non-Apple voices as it's pretty accurate.
http://ipa-reader.xyz <http://ipa-reader.xyz/>
—
Reply to this email directly, view it on GitHub <https://gist.github.com/e1284682216d4e4a90fa474a676fb453#gistcomment-4331749>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAL5QCPXAXB2TW4ZADJYWETWCWSC5ANCNFSM5T7LAM4A>.
You are receiving this because you commented.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
This code is a bit older and I found a few bugs since. Mainly if you have a different region set on your phone the IPA might sound very strange. The latest version that works better is included in this package > OEVoice