If You Want To Sing Out…

I invited my husband, Barry, to be a guest blogger for this piece, because I can’t begin to explain how he achieved this miracle. Please note: Barry is a Principal Software Engineer by day. I will attempt to translate anything I think may trip up us lay people. 

Jessie did some “phrase banking” when she was first diagnosed with ALS in 2017 in anticipation of losing her voice. She recorded a library of phrases that are stored for verbatim use on a text to speech device. When she types “Hello!” the device plays her banked recording of “Hello!” in her own voice. 

Jessie intended to do “voice banking” as well, in which she records a huge quantity of sentences, and then a company generates a synthesized voice that sounds like Jessie’s original voice. Unfortunately, between teaching and looking for a new house, Jessie only got halfway through recording the sentences before her speech was affected.  

Like most folks in this situation, she listened to the various canned voices on her Tobii Dynavox device and chose a female voice that sounded “close enough”. These voices are good. The technology has come a long way from the computerized voices we heard growing up (…think “War Games” or Stephen Hawking’s early use of text to speech technology) but it was never her voice.

Well, we recently heard an interview with Eric Schmidt, former CEO at Google. He was talking about how they had a prototype a while back that could create a realistic voice clone from no more than a few minutes of audio. It was not released at the time for various reasons but it got me thinking that the time was right for us to return to this issue and do something!

As you know, Jess was a public school teacher for 24 years and she had a number of classroom videos she filmed. I extracted a bunch of clear voice audio from these and fed them into the ElevenLabs voice cloning app (https://elevenlabs.io/).

ElevenLabs markets themselves primarily for agentic (autonomic AI) content creation: voice overs, audiobooks, podcasts, chat bots, etc. However, their cloning capabilities were reviewed as highly accurate and expressive.

Indeed, the voice modelling was spot on from the ~60 minutes of input we provided. We now had a nearly perfect sounding Jess-bot!

I thought at this point that we might use the voice clone to generate content that could be fed into the traditional voice banking apps, having the voice clone do the “recordings” that you would normally do yourself in this process. Unfortunately, there is a maze of intentional licensing barriers to such an approach.

However, ElevenLabs does have a partnership with BridgingVoice which is a non-profit organization focused on ALS and assistive communication technology. “We help people living with ALS (pALS) maintain their ability to meaningfully communicate with family, caregivers, and medical professionals, no matter the stage of their disease.” BridgingVoice picked up the licensing fees for our use of the ElevenLabs voice clone API (Application Programming Interface). (https://bridgingvoice.org)

Then I wrote a little application to act as a broker between the various components. It takes in the input text from her typing app, communicates with the ElevenLabs API to vocalize that text, and processes and plays the audio returned. (From Jessie: if this paragraph confuses you, just go on to the next one. It won’t affect your understanding.)

And voila, Jess has her voice back! She types (with her eyes) anything she wants to say and it is vocalized in near real time with her cloned voice.

Note that BridgingVoice has a Tobii Dynavox integration app similar to the one I wrote that they provide free of charge to the pALS they support. If you need this, you should totally use it – it works just fine. I decided to write my own software because…

1. I wanted to learn more about the technology and I was already halfway down the road of thinking through various issues in my head, so I couldn’t stop myself. 🙂
   
2. The BridgingVoice app had some volume integration issues I really wanted to address to make the user experience smoother.
   
3. I’m sure the ElevenLabs Text to Speech API is going to evolve quickly and I wanted to be in a position to incorporate new features without annoying the BridgingVoice folks with lots of questions and feature requests.
   
4. I also added a bunch of additional INFO and DEBUG diagnostic logging so I can help Jess when things aren’t working so well. (From Jessie: I ignore the technical terms I don’t understand, if it doesn’t get in the way of my overall understanding.)
   2026-02-22 13:41:04.202 [INF] VoiceApp Main()…
   2026-02-22 13:41:04.234 [INF] Text: 86 characters
   2026-02-22 13:41:04.867 [INF] GetAudio(): 0.627 seconds
   2026-02-22 13:41:10.624 [INF] PlayAudio(): 5.754 seconds
   2026-02-22 13:41:10.628 [INF] VoiceApp Main(): 6.516 seconds

The “Speak” button to the right on her eye gaze keyboard vocalizes with the same canned voice as before. The new “Voice” button on the left vocalizes with the clone. But the clone is dependent on a network connection to access the ElevenLabs API so we left the Speak button in place as a backup when we’re out of the house, or having a Wi-Fi outage, or whatever.

(Also, the clone tends to add confusing emotional tones to texts with just a word or two in length. So the old voice or banked phrases are still much better at saying “Yeah” or other simple texts that don’t provide enough language context for the clone to model very well.)

This is a small clip of Jessie from 2011 leading a class of high schoolers through a reading of Homer’s Odyssey:

And this is Jessie 2.0 today:

(Ok. You got me. I typed that, not Jess. But you get the idea.)

I’m sure this process is going to get even easier from here on out. It’s not hard to imagine a time when you turn on your text to speech device for the first time and it prompts you to record or input a few simple voice prompts. And that’s it, you have a highly accurate personalized digital voice. I’m just glad we were able to manually connect the tech dots for Jess’s benefit today.

So let Jessie know if you’d like to come by for a visit some time soon. She’d be happy to catch up and talk with you in her own unique and beautiful voice.

…Sing Out.

– barry