Apple recently introduced Personal Voice to newer devices on various hardware in their lineup. I have had a little experience with the basic concept behind this sort of technology from my time at Microsoft where I dabbled with one of Microsoft’s Azure cognitive services to do something similar.
The basic concept behind these experiences is that you record some set of known text and then software converts that into a synthetic version of your voice. In Apple’s case it is 150 phrases ranging from just a few words to maybe at most 20 words in a single phrase.
After you finish recording, there is some processing time and then your voice is ready to use. On an iPhone 15 Pro, my voice was ready in about five hours. You are not able to do anything else with the phone while this is happening. On an M1 MacBook Air from 2020, processing took about two hours and I was able to do other tasks at the same time, such as writing this blog post.
Once your voice is created, you can use it as one of the voices available with Apple’s Live Speech feature. This allows you to type in various apps where you would typically use your voice and have the synthetic voice used. It compliments the existing voices Apple makes available and has the added benefit of allowing you to have some relationship to your own voice used in the experience. In situations where a person may know that they are going to lose their voice ahead of time, it does offer some ability to preserve your own speech.
Multiple factors influence the quality of the end result here—Microphone, recording environment and more, just to name a few. For short phrases it likely is not noticeable but in my sample, even the pace at which I appear to have read the samples was different. There is a 21 second difference in having the voice read back the same text.
I made two voices in trying this experience. The first was recorded using the default Apple headphones on an iPhone 15 Pro. The second using an Arctis 7 headset. Both samples are my Apple Personal Voice reading my blog post on Accessibility Island.
I have also made a sample of my original voice sample of three phrases and then Apple’s Personal Voice speaking those phrases from my recording with the Arctis 7 device. The Personal Voice speaking the phrases is the result of my typing them into an edit box and asking for them to be spoken using my newly created voice. The phrases are in this recording and have the original voice sample followed immediately by the Personal Voice speaking the phrase. After all three phrases are played, the entire series is duplicated once. The phrases are:
can you call me in an hour
Did you remember to take out the trash?
Is she going to the grocery store now or in the morning?
Creating a personal voice is straight forward. On whatever device you are using, go to Settings:Accessibility:speech:Personal Voice. You’ll be prompted to record a short phrase to test your recording environment and advised of any changes you should make, such as too much background noise. You then start the process of recording 150 phrases. They do not all need to be recorded at once. When you are finished, you’ll be advised to lock your phone if doing this on an iPhone or just ensure your computer is charged if using a Mac.
When the voice is created, you can start using it with Live Speech by going to the same Speech area of Accessibility settings and going into Live speech. Turn Live Speech on and then pick from the list of voices. Your personal Voice should be listed.
If you are doing all of this with VoiceOver, Apple’s screen reader, as I did, the process of creating a voice works well with VoiceOver. You can use VoiceOver to read the phrase to be read, then activate a record button and repeat the phrase. Recording stops when you stop speaking. If you turn on a setting for continuous recording, you will advance to the next phrase automatically and can repeat the process. I did notice that sometimes VoiceOver automatically read the next phrase but not always. Focus seems to go to the Record button and I suspect there is a timing issue between the phrase being spoken and VoiceOver announcing the newly focused button.
Having created two voices, I would say it is probably a good idea to take a short break during the reading of the 150 phrases from time to time. I found myself not speaking as clearly as I wanted once in a while as well as having sort of the same singsong phrasing. Listening to my voice samples and how the voice came out, I would also say the microphone used has a big impact on the voice quality. This isn’t surprising but is made apparent to me comparing the samples of what my recordings sounded like and how that turns out when the same text is spoken by Personal Voice. I don’t think either microphone that I used would be what I would recommend for creating a voice to be used permanently.
I was curious if Apple would allow the personal voice you create to be used with VoiceOver. I didn’t expect it would be possible and that does seem to be the case.
As with pretty much anything in AI, synthetic speech is a rapidly changing technology. There are certainly higher quality voices in the arena of synthesized speech but Apple has done a good job at allowing you to tap your own voice on consumer hardware in an easy to use process. Listening to my own voice, it is clear it isn’t me and I wasn’t expecting it to be. But even on the basic hardware I used, there are characteristics of my voice present and if I were in a situation where I was going to lose my physical voice permanently, this is one option I would definitely explore further.
Of course, now in iOS 18,to use personal voice is available with voiceover. It’s no 11 labs, that’s for sure. As for your examples, I liked the Arctis headset much better than the Apple as far as crispness and clarity. What was interesting was the entire pitch of the sample seemed to be higher with the second headset, and I’m not sure why that would be. If you ran both at the same time, you would sound like two different people using the same intonation. I hope that you can revisit this topic and try continuous recording again to see if they have improved the process. Maybe it would work better with the Eloquence voice. I wasn’t able to find a cancel button in the player so the only way to close the player was to hit the back button. I was also looking for a link to sign up to receive your entries by email on the page and could not find one even after opening the menu, but that could be my own incompetence. All that not withstanding, nice job, and thanks for taking the time.