Ask HN: My wife might lose the ability to speak in 3 weeks – how to prepare?

Hey, speech ML researcher here. Make sure you have different recordings of different contexts. fifteen.ai&#x27;s best TTS voices use ~90 min of utterances, some separated by emotion. If you&#x27;re having her read a text, make sure it&#x27;s engaging--we do a lot of unconscious voicing when reading aloud. Tbh, if she has a non-Anglophone accent, you&#x27;re going to need more because the training data is biased towards UK&#x2F;US speakers.<p>If you want to read up on the basics, check out the SV2TTS paper: <a href="https:&#x2F;&#x2F;arxiv.org&#x2F;pdf&#x2F;1806.04558.pdf" rel="nofollow">https:&#x2F;&#x2F;arxiv.org&#x2F;pdf&#x2F;1806.04558.pdf</a> Basically you use a speaker encoding to condition the TTS output. This paper&#x2F;idea is used all over, even for speech-to-speech translation, with small changes.<p>There&#x27;s a few open-source version implementations but mostly outdated--the better ones are either private for business or privacy reasons.<p>There&#x27;s a lot of work on non-parallel transfer learning (aka subjects are saying different things) so TTS has progressed rapidly and most public implementations lag a bit behind the research. If you&#x27;re willing to grok speech processing, I&#x27;d start with NeMo for overall simplicity--don&#x27;t get distracted by Kaldi.<p>Edit: Important note! Utterances are usually clipped of silence before&#x2F;after so take that into account when analyzing corpus lengths. The quality of each utterance is much much more important than the length--fifteen.ai&#x27;s TTS is so good primarily because they got fans of each character to collect the data.

My mom lost her ability to speak, and what you are going to find is that your life and how you interact with everyone will have to change. Human verbal communication is very fast. She will find it difficult to be part of normal conversations. Without lots of help, she will start to fade into the background of conversations, because she can&#x27;t keep up. You will have to help her be a part of things. It will be a depressing experience for her, and you will have to help her. People will look at her differently like she is mentally handicapped. (I know she won&#x27;t be, but people will assume that she is even unconsciously). I recommend finding her a therapist if she has to go through this transition.

Record her reading the texts of a standardized text training corpus.<p>That way, you can retrain an existing AI to do text to speech with her own voice.<p>Edit: here&#x27;s a link to the corpus that I believe Mozilla uses <a href="http:&#x2F;&#x2F;www.openslr.org&#x2F;12&#x2F;" rel="nofollow">http:&#x2F;&#x2F;www.openslr.org&#x2F;12&#x2F;</a>

I would also suggest looking into learning American Sign Language (of course alongside this project). While communicating via keyboard is workable and good for communicating with the wider world, ASL would be much more convenient for communicating between you two -- and a very interesting language to boot. It is a foreign language thats not related to English besides a few loan words, but there&#x27;s tons of online resources and most universities have classes as well. Plus, you also can experience beautiful Deaf culture, with a rich storytelling and poetic tradition that blends language, gesture, acting, and pantomime in a way thats just impossible to translate to a spoken language.<p>The downvoted commenter was being a jerk, but I do think learning ASL is an option worth looking into.

I have a progressive neurodegenerative disease and lost most my ability to speak about 3 years ago. What you are proposing is super cool, but you might be overthinking this. These things (text to speech, etc) are more awkward than practical in real life. Also, make sure your wife is completely on board. Seeing old clips and hearing my voice is actually kind of depressing to me. Here is my actual advice:<p>Outside of social situations, it honestly hasn&#x27;t been that big of deal for me. As a remote developer, my job has remained the same. My managers and co workers have been super supportive. I send messages during meetings to one person who will read it aloud for me.<p>With text and social media, I still keep up with friends and family. Most medical appointments, etc, can be made online. SprintIP relay is free for deaf&#x2F;speech impaired, and it allows the caller to type what they want to say and a representative will relay this to the other party. It works via the web or a mobile app. <a href="https:&#x2F;&#x2F;www.sprintrelay.com&#x2F;sprintiprelay" rel="nofollow">https:&#x2F;&#x2F;www.sprintrelay.com&#x2F;sprintiprelay</a><p>Banks, brokers, or anything involving personal info (like SS#) usually requires a voice phone call. I have my wife call and explain the situation. I can whisper yes, as they occasionally require me to give permission. Some call center representatives have no idea how to handle this situation, and will just stick to the script saying they have to speak to me the entire time. My wife just thanks them, calls back, and hopes for someone more understanding.<p>There are awkward encounters where people don&#x27;t know you can&#x27;t speak, and will respond by speaking louder and slower. These people will also assume you are not intelligent and be dismissive. This is just one of the things you have to deal with.<p>I sincerely hope the procedure goes well and you wife doesn&#x27;t have to deal with this. Just know that even if the worse happens, she can have a normal and productive life!

We get quite a few requests for this at Resemble (<a href="https:&#x2F;&#x2F;resemble.ai" rel="nofollow">https:&#x2F;&#x2F;resemble.ai</a>). We can get her to record right on our website or you can upload an existing file (along with a video of her consent) on the platform. Feel free to shoot me a message and I&#x27;d be happy to help build a voice for her.

I don&#x27;t know if you have kids&#x2F;grandkids&#x2F;nieces or nephews (or plan to have those) but it might be nice to record your wife reading some books out loud.<p>Not only will you have your own personal &quot;audio books&quot; of Harry Potter&#x2F;The Hobbit&#x2F;Chronicles of Narnia&#x2F;Oi Frog&#x2F;Alice in Wonderland&#x2F;Roald Dahls etc etc for any kids&#x2F;grandkids&#x2F;relatives etc that will hopefully be something treasured in its own right, but you&#x27;ll also have a large corpus of training data from well-known texts that you can retrain over and over as the tech improves in the future. Might be worth chucking in some other well-known texts to avoid over-fitting on a &quot;kids&#x27; story voice&quot; - maybe something plain like inauguration speeches&#x2F;declaration of independence&#x2F;magna carta&#x2F;etc.<p>Obviously I&#x27;d focus on gathering raw material now, and focus on the reconstruction later when you&#x27;ve all recovered mentally and physically to whatever happens. The more data the better when it comes to this sort of thing. There might not be something &quot;simple&quot; right now (e.g. you could probably implement the WaveNet or similar paper yourself today, and training it up on some GPUs in your spare room etc, but in a few years there might be a nice WYSIWYG&#x2F;SaaS thing for it), but with the recordings safely stored you&#x27;ll obviously be able to use it in the future.<p>Best of luck to you both.

I don&#x27;t have any answers to give you, but I want to say that this is a really loving and beautiful thing you&#x27;re trying to do.

Other resources you may want to explore are r&#x2F;mute and r&#x2F;deaf subreddits. Both also have Discord servers listed in the sidebars.<p>Having spent a good deal of time in hospitals, a few things I recommend... 10’ phone cable since outlets can sometimes be far from the bed, cheap slippers she can wear to walk around (stepping in a hospital hallway mystery puddle wearing just socks is very unpleasant), comfy clothes that you don’t mind having ruined (T-shirts, underwear, shirts, pajama pants - they can temporarily unhook the IV so she can put a T-shirt on), earplugs, eye mask. If she’s going to be on liquid-only diet, bring your own since hospital food is not great, not terrible. Soylent&#x2F;Orgain&#x2F;Ensure if she’s permitted that, otherwise good quality Italian ices are such a nice treat and most hospitals have a patient fridge&#x2F;freezer you can store them in. Broth, but go to a restaurant or grocery store&#x2F;farmers market with hot soup bar and fill a container with just the broth from the chicken noodle soup. It’s INFINITELY better than boxed broth.<p>Hopefully all of your research and preparation will be for nothing, I wish you and your wife a successful surgery!

Speech-language Pathology student here. I would recommend going to see a speech therapist. It will likely be covered by your health insurance. Find an SLP who specializes in AAC (Augmented and Alternative Communication) who can help your wife communicate if she loses her speech. Your DIY approach could work, but having support from an SLP to help her learn the system, and come up with other options if it doesn&#x27;t cover all of her communication needs, will go a long way.

Just have her carry a good microphone at all times to record everything she says until that point, to have a maximum amount of samples. If you can&#x27;t &quot;deepfake&quot; it today, maybe you will be able to do it tomorrow, but at least you will have the data.

Others here are addressing technical solutions, but I don&#x27;t see anyone here covering non-verbal communication. IMO, that&#x27;s going to be just as important.<p>I am going to assume that your wife and you have a healthy relationship with strong communication, in part because you&#x27;ve developed an intuition for her body language and other non-verbal communication methods. In the scenario where she loses her ability to speak, even if she happily and completely takes to whatever technical solution(s) you offer to replace that, I think it&#x27;s likely she will reflexively lean more heavily on those non-verbal channels, and you&#x27;re going to need to get better at reading them than you are now.

This might get you started:<p><a href="https:&#x2F;&#x2F;speech.microsoft.com&#x2F;customvoice" rel="nofollow">https:&#x2F;&#x2F;speech.microsoft.com&#x2F;customvoice</a><p>I imagine if MS offers custom voices then the other text to speech providers do as well.<p>Good luck

Some (decades old) research on this involved a research team creating a video of JFK saying &quot;I never met Forrest Gump&quot;. I found a writeup in Google Books: <a href="https:&#x2F;&#x2F;books.google.com&#x2F;books?id=mQtGVQeQplcC&amp;pg=PA208&amp;lpg=PA208&amp;dq=%22I+never+met+forrest+gump%22&amp;source=bl&amp;ots=k3PobhFWaY&amp;sig=ACfU3U3VlGf4aIdU1Q_JRllhb8AwVNzeLA&amp;hl=en&amp;sa=X&amp;ved=2ahUKEwiT7dyxkvrpAhUrHDQIHVUfDwQQ6AEwAHoECAoQAQ#v=onepage&amp;q=%22I%20never%20met%20forrest%20gump%22&amp;f=false" rel="nofollow">https:&#x2F;&#x2F;books.google.com&#x2F;books?id=mQtGVQeQplcC&amp;pg=PA208&amp;lpg=...</a><p>&gt; We evaluated our Kennedy results qualitatively along the following dimensions: ... naturalness of the composited articulation; ...<p>Obviously the state of the art will have advanced, but maybe this can point the way toward more current research.<p>While I tend to agree with everyone else that this <i>can be</i> a great idea, my instinct is to float the idea to your wife first and see how she responds. I can imagine someone taking this negatively.

If she loses her ability to speak there are many ways to help her out, but nothing can replace the sound of her voice, especially for those important moments.<p>Just in case. Record specific messages for various people in her life, that can be used repeatedly, Children, Mom, Dad, siblings, in-laws, friends, messages like: &quot;X, I love you&quot;, &quot;X, I miss you.&quot;, &quot;Mommy loves you!&quot; &quot;Give me a hug&quot;. &quot;Holiday Greeting&quot;, &quot;Happy Birthday&quot;,&quot;I&#x27;m so proud of you!&quot; favorite happy saying, frustration saying,<p>You get the idea.

What about recording messages to other people for future events (e.g. graduation of a child, birth of grandchild etc.)?<p>Recording a message to a yet unborn grandchild is maybe something we could all do!

When my dad lost his speech, we had Boogie Board Jot devices all over the house. It made writing short notes and simple dialogs much less tedious.<p>We also used the Verbally premium iPad app to help give him a voice and make transactions on easier.<p>Wishing you all the best.

The paper &quot;Generalization Of Audio Deepfake Detection&quot; gives an overview.<p>The paper <a href="https:&#x2F;&#x2F;arxiv.org&#x2F;abs&#x2F;1904.05441" rel="nofollow">https:&#x2F;&#x2F;arxiv.org&#x2F;abs&#x2F;1904.05441</a> has a list of spoofing methods.<p>Here&#x27;s one method as paper <a href="https:&#x2F;&#x2F;arxiv.org&#x2F;pdf&#x2F;1806.04558.pdf" rel="nofollow">https:&#x2F;&#x2F;arxiv.org&#x2F;pdf&#x2F;1806.04558.pdf</a><p>And here on GitHub <a href="https:&#x2F;&#x2F;github.com&#x2F;CorentinJ&#x2F;Real-Time-Voice-Cloning" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;CorentinJ&#x2F;Real-Time-Voice-Cloning</a>

For an open-source approach, the MaryTTS project has a guide on how to add new voices to their tool: <a href="https:&#x2F;&#x2F;github.com&#x2F;marytts&#x2F;marytts&#x2F;wiki&#x2F;VoiceImportToolsTutorial" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;marytts&#x2F;marytts&#x2F;wiki&#x2F;VoiceImportToolsTuto...</a>

You may want to look up what was done for Roger Ebert. He has lost his voice due to surgery, but because of the vast corpus of audio recordings of him, a viable text to speech engine was able to be created.<p>It’s a bit dated at this point, but I imagine the research has vastly improved since then.<p>It’s a very good question though. A decade ago this was able to be done for one man. Is it now possible to be done for anyone? Like others, I’d guess the first step is to record everything while you can.

I wrote <a href="https:&#x2F;&#x2F;trumped.com" rel="nofollow">https:&#x2F;&#x2F;trumped.com</a><p>You ideally want five hours of clean speech (good microphone, no background noise, high sample rate). It should be spoken clearly, in a single tone or mood. My model sounds awful because the data isn&#x27;t consistent, and the room tone and microphones are terrible.<p>If you want different prosody or moods, don&#x27;t mix them in the same data set.<p>You can experiment with transfer learning LJSpeech with Nvidia Tacotron2 right now. Glow-tts is also promising.<p>You&#x27;ll start to get results with fifteen minutes of sample data, but for high quality you want a lot of audio.<p>Have your wife read a book and record it. The training chunks will be ~10 seconds apiece, so keep that in mind for how to segment the audio.<p>Focus on getting lots of good sounding data. Hours. The models will improve, but this may be your only shot of acquiring the data.<p>Download the LJSpeech dataset and listen to it. See how it sounds, how it&#x27;s separated. That is a fantastic dataset that has yielded tremendous results, and you can use it for inspiration.

Here&#x27;s a simple and practical solution:<p>Get a decent audio headset, have it record the audio to her phone, and spend hours talking to her about whatever. Preferably in a reasonably quiet environment.<p>Just spend a lot of time talking. You don&#x27;t have to talk to her through a headset. Just make sure hers is recording her voice.<p>It would be easy, painless, and probably good for the relationship too.

At a minimum get the following list of sentences recorded in her voice, <a href="http:&#x2F;&#x2F;www.festvox.org&#x2F;cmu_arctic&#x2F;cmuarctic.data" rel="nofollow">http:&#x2F;&#x2F;www.festvox.org&#x2F;cmu_arctic&#x2F;cmuarctic.data</a><p>Make sure the recordings are of a good quality. This will ensure that you will have a baseline TTS of her voice at the minimum.

(off topic) Record a few things for her future self. E.g. favourite quotes, frequently used phrases.

Make sure to record with the best microphone you can find and in the quietest room you can find. Makes a huge difference in the resulting TTS.

You might look at resources for ALS patients.<p>Since ALS (aka Lou Gehrig&#x27;s disease) is a degenerative motor neuron disease, people with ALS can pretty much count on eventually losing the ability to speak. So &quot;voice banking&quot; is apparently pretty common.

Not exactly what you&#x27;re asking for, but I wrote an app for this scenario:<p><a href="https:&#x2F;&#x2F;play.google.com&#x2F;store&#x2F;apps&#x2F;details?id=org.anaisbetts.sirene" rel="nofollow">https:&#x2F;&#x2F;play.google.com&#x2F;store&#x2F;apps&#x2F;details?id=org.anaisbetts...</a><p>This is a text-to-speech app with a very keen emphasis on <i>Day To Day</i> usage - the UX will put the focus at the right places, help you reply faster, etc. I used it for a full month when I was unable to speak after voice surgery and it made a big difference, other folx have reported the same

This is probably a really stupid suggestion but just in case.<p>Do you and your wife drink alcohol a bit? If so might it be worth having a couple of drinks in a quiet setting with her one evening with microphones running? I&#x27;m not suggesting getting wasted! I&#x27;m just wondering whether it might help to catch her getting more animated or &quot;natural&quot; in conversation. I was thinking this might help make the resulting synthesized speech capture even more of her personality than reading children&#x27;s books or subsets of AI corpora etc.

I have had good results with this. <a href="https:&#x2F;&#x2F;www.resemble.ai&#x2F;" rel="nofollow">https:&#x2F;&#x2F;www.resemble.ai&#x2F;</a> It is based on this open source work. If you want to run it yourself. <a href="https:&#x2F;&#x2F;github.com&#x2F;CorentinJ&#x2F;Real-Time-Voice-Cloning" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;CorentinJ&#x2F;Real-Time-Voice-Cloning</a><p>The voice cloning can be done in a matter of minutes. (&lt; an hour) Its also very easy to use the website.<p>Best of luck!

Strongly suggest reaching out to Dr. Rupal Patel (<a href="https:&#x2F;&#x2F;www.linkedin.com&#x2F;in&#x2F;rupalvocalid" rel="nofollow">https:&#x2F;&#x2F;www.linkedin.com&#x2F;in&#x2F;rupalvocalid</a>) of Northeastern University (<a href="https:&#x2F;&#x2F;coe.northeastern.edu&#x2F;people&#x2F;patel-rupal&#x2F;" rel="nofollow">https:&#x2F;&#x2F;coe.northeastern.edu&#x2F;people&#x2F;patel-rupal&#x2F;</a>) and VocaliD (<a href="https:&#x2F;&#x2F;vocalid.ai&#x2F;about-us&#x2F;" rel="nofollow">https:&#x2F;&#x2F;vocalid.ai&#x2F;about-us&#x2F;</a>). She&#x27;s a licensed Speech-Language Pathologist (<a href="https:&#x2F;&#x2F;web.northeastern.edu&#x2F;cadlab&#x2F;publications&#x2F;RupalPatel_CV_WEB.pdf" rel="nofollow">https:&#x2F;&#x2F;web.northeastern.edu&#x2F;cadlab&#x2F;publications&#x2F;RupalPatel_...</a>) and she and her husband, Dr. Deb Roy, did the Human Speechome project (<a href="https:&#x2F;&#x2F;en.wikipedia.org&#x2F;wiki&#x2F;Human_Speechome_Project" rel="nofollow">https:&#x2F;&#x2F;en.wikipedia.org&#x2F;wiki&#x2F;Human_Speechome_Project</a>). She was also my doctoral advisor and I feel confident saying she would be very interested in talking with you.

Do you have children? Perhaps - record her reading a few favorite children&#x27;s books.

ML will require a lot of samples for getting it as desired. I will say, let your wife carry an attached microphone and meet all the people she wishes to talk at least once. collect all the audio data, and you can use it later. &lt;ake all the available moments memorable for her like If you have child record a message from your wife for next 10 birthday of child.

Consider investing in a good microphone for recording. A Blue Yeti is ~$200.

Not to discourage you from making voice recordings and all that, but as someone who is handicapped and sometimes has trouble speaking because of it:<p>1. I spend a lot of time online. It doesn&#x27;t matter so much there. I do a lot of typing.<p>2. My oldest son, who had serious output difficulties as a child, is talented at inferring what I need from a gesture and a grunt. This has proven enormously helpful.<p>3. Consider using her phone as a communication device. It&#x27;s small and people tend to take their phone everywhere and she can type out what she wants to say.<p>4. Writing tweets can help a person learn to say things more succinctly. I do freelance writing and figuring out how to say things succinctly is a talent you can develop. (It&#x27;s something I have to work at -- I&#x27;m a &quot;would have written you a shorter letter if I had more time&quot; type of person.) This can help enormously when you face communication barriers.<p>5. Take some time to deal with the emotional stuff. It matters.<p>I&#x27;m sorry you are facing this. Best of luck.

Here&#x27;s a recent work [0] where you can train the model with 10s audio and convert any &quot;text to speech&quot; (all doable in the browser). I tried with Google Colab demo [1] and its performance fluctuates with the training audio sample that you give it so might need some trial and error to get the sweet spot.<p>Also the model is not saved in the browser with Colab so you might also want to do it locally to save it eventualy (if it comes to that).<p>All the best mate!<p>[0] Main repo: <a href="https:&#x2F;&#x2F;github.com&#x2F;CorentinJ&#x2F;Real-Time-Voice-Cloning" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;CorentinJ&#x2F;Real-Time-Voice-Cloning</a> [1] Google colab repo to try it out: <a href="https:&#x2F;&#x2F;github.com&#x2F;CorentinJ&#x2F;Real-Time-Voice-Cloning&#x2F;blob&#x2F;master&#x2F;demo_toolbox_collab.ipynb" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;CorentinJ&#x2F;Real-Time-Voice-Cloning&#x2F;blob&#x2F;ma...</a>

Hi, I like your idea for your wife. Hope the surgery will succeed without damage to her speaking. I&#x27;m from Nvidia and know well the team behind NeMo toolkit. Happy to connect you to the team if that helps. You may send me an email to ardenwood.bruin_at_gmail.com. -- Michael

This may also be useful. Free and open source.<p><a href="https:&#x2F;&#x2F;www.tobiidynavox.com&#x2F;en-gb&#x2F;software&#x2F;web-applications&#x2F;message-banking-2&#x2F;" rel="nofollow">https:&#x2F;&#x2F;www.tobiidynavox.com&#x2F;en-gb&#x2F;software&#x2F;web-applications...</a>

The mycroft voice assistant has some tooling they used to create voices.<p><a href="https:&#x2F;&#x2F;mycroft.ai&#x2F;blog&#x2F;mimic-2-is-live&#x2F;" rel="nofollow">https:&#x2F;&#x2F;mycroft.ai&#x2F;blog&#x2F;mimic-2-is-live&#x2F;</a> <a href="https:&#x2F;&#x2F;github.com&#x2F;MycroftAI&#x2F;mimic2" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;MycroftAI&#x2F;mimic2</a><p>Search Results Web results<p>Festival Speech Synthesis has a tool for recording speech databases, and some tutorials for training festival voices. <a href="http:&#x2F;&#x2F;www.cstr.ed.ac.uk&#x2F;research&#x2F;projects&#x2F;speechrecorder&#x2F;" rel="nofollow">http:&#x2F;&#x2F;www.cstr.ed.ac.uk&#x2F;research&#x2F;projects&#x2F;speechrecorder&#x2F;</a>

You need to do voice banking. It is imperative that you do so, so that your wife keeps her identity no matter what.<p>What you need to do is spend the entire next 3 weeks doing voice banking. This will give your wife a text-to-speech voice (SAPI 5 voice, or others, for example). You record phrases that the voice banking service wants you to speak, with a <i>high quality headset (best if wired) in a quiet setting</i>.<p>The more sentences (samples) you have, the better the voice will be, obviously. But, there are services out there that will update the recordings, as the technology gets better, and that is the way to go, in terms of choosing the &quot;best service&quot;.<p>The voice banking services that people typically use are here: <a href="https:&#x2F;&#x2F;www.mndassociation.org&#x2F;professionals&#x2F;management-of-mnd&#x2F;aac-for-mnd&#x2F;voice-banking&#x2F;equipment-and-services&#x2F;" rel="nofollow">https:&#x2F;&#x2F;www.mndassociation.org&#x2F;professionals&#x2F;management-of-m...</a><p>I would say that Acapela my-own-voice is currently the best technology. Obviously there are open source technologies, but you do not have the luxury of time to figure all of that out. However, you should do your own voice banking for later post-processing on your own with open source stuff.<p>There is also a free version of voice banking available, but I would only recommend it as a secondary tool: <a href="https:&#x2F;&#x2F;www.modeltalker.org&#x2F;" rel="nofollow">https:&#x2F;&#x2F;www.modeltalker.org&#x2F;</a><p>This app (iOS and Android) for example, allows you to use your personal voice banked text-to-speech voice, to talk: <a href="https:&#x2F;&#x2F;therapy-box.co.uk&#x2F;predictable" rel="nofollow">https:&#x2F;&#x2F;therapy-box.co.uk&#x2F;predictable</a><p>This is another great app that allows you to use your personal voice banked text-to-speech voice: <a href="https:&#x2F;&#x2F;www.assistiveware.com&#x2F;products&#x2F;proloquo4text" rel="nofollow">https:&#x2F;&#x2F;www.assistiveware.com&#x2F;products&#x2F;proloquo4text</a><p>Source: Disabled engineering student, who is extremely interested in assistive technology. I would love to be a rehabilitation engineer.

It might also be worth recording normal conversations you have around the house as a fallback. You can always cut it up later and feed it into these systems.<p>Best of luck to the two of you. I really hope you don&#x27;t ever need this technology.

You might want to try DIY&#x27;ing something like this [1] depending on the extensiveness of her surgery. It basically records electrical signals (EMG) emitted by the vocal chords (subvocalizations) and can convert it to text with ML&#x2F;other signal processing algorithms. Basically a rudimentary version of the transhumanist Brain-Computer Interfaces that would enable telepathy.<p>[1] <a href="https:&#x2F;&#x2F;dam-prod.media.mit.edu&#x2F;x&#x2F;2018&#x2F;03&#x2F;23&#x2F;p43-kapur_BRjFwE6.pdf" rel="nofollow">https:&#x2F;&#x2F;dam-prod.media.mit.edu&#x2F;x&#x2F;2018&#x2F;03&#x2F;23&#x2F;p43-kapur_BRjFwE...</a>

This can be trained using only 5 Seconds of reference audio: <a href="https:&#x2F;&#x2F;google.github.io&#x2F;tacotron&#x2F;publications&#x2F;speaker_adaptation&#x2F;" rel="nofollow">https:&#x2F;&#x2F;google.github.io&#x2F;tacotron&#x2F;publications&#x2F;speaker_adapt...</a> <a href="https:&#x2F;&#x2F;arxiv.org&#x2F;pdf&#x2F;1806.04558.pdf" rel="nofollow">https:&#x2F;&#x2F;arxiv.org&#x2F;pdf&#x2F;1806.04558.pdf</a><p>It&#x27;s been mentioned a bit already, but thought it was worth calling out. This may be one of the lowest-overhead ways to start experimenting, at least in terms of data collection.

Your approach towards the situation might determine the life quality of you and your wife. I can&#x27;t imagine how it&#x27;s like to think in a logic way while you&#x27;re in the middle of such of an emotional event.

<a href="https:&#x2F;&#x2F;dathudeptrai.github.io&#x2F;TensorflowTTS&#x2F;" rel="nofollow">https:&#x2F;&#x2F;dathudeptrai.github.io&#x2F;TensorflowTTS&#x2F;</a> is the state of the art and feels natural enough

First off I&#x27;m sorry you&#x27;re going through that, it sounds really tough. We sometimes have families use us for this (<a href="https:&#x2F;&#x2F;trysaga.com" rel="nofollow">https:&#x2F;&#x2F;trysaga.com</a>) as a way to collect voice recordings of loved ones, to record and share a large number of memories and stories in their voice and have them saved forever. You can download all the recordings to keep. It&#x27;s free right now and I&#x27;d be happy to help out and make sure it got you what you needed, let me know.

Probably not what you were seeking, but I have to imagine it would be similar to long periods I have spent in a non-verbal state. Being allowed to exist and just smile or laugh as a &quot;part&quot; of the conversation around me was like sunlight on a dark day. The range of human emotion and expression often overlaps enormously between people. Sometimes pretending you&#x27;re voice is really the good you hear around you and not the throat mumblings that cause so much conflict is the most beautiful dream.

Sorry about this. Record as much voice as you can now (stereo too?), then you’ll have time to find the right solution and improve it as the technology gets better in time

I went through something similar with a parent years and years ago. I wanted to be able to do things to help with what would eventually be lost.<p>I have to say I didn&#x27;t help as much as I thought I could and afterwards I was always wondering if I could have used this technology or that and done more.<p>So - I think you should recognize that you can only do so much, we&#x27;re doing the best we can, and in the end we are all winging it.

You might contact the following company: <a href="https:&#x2F;&#x2F;www.acapela-group.com&#x2F;solutions&#x2F;acapela-voice-factory&#x2F;" rel="nofollow">https:&#x2F;&#x2F;www.acapela-group.com&#x2F;solutions&#x2F;acapela-voice-factor...</a><p>There&#x27;s also open source TTS from Mozilla: <a href="https:&#x2F;&#x2F;github.com&#x2F;mozilla&#x2F;TTS" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;mozilla&#x2F;TTS</a>

Hope I am not repeating any comments here. My suggestion is that you start recording as soon as possible and as much as possible without worrying about technicalities. You can also use if you have any old voice records or videos with a relatively good voice quality. For now maybe she can read a book aloud in a silent room. After you have the data I can also help if you like to create a TTS model.

This is not open source but this was very good from their demo in terms of your own voice reproduction.<p><a href="https:&#x2F;&#x2F;www.descript.com&#x2F;lyrebird-ai" rel="nofollow">https:&#x2F;&#x2F;www.descript.com&#x2F;lyrebird-ai</a><p>I hope good folks in there will help you, try reaching them.<p><a href="https:&#x2F;&#x2F;m.youtube.com&#x2F;watch?v=VnFC-s2nOtI" rel="nofollow">https:&#x2F;&#x2F;m.youtube.com&#x2F;watch?v=VnFC-s2nOtI</a>

I&#x27;ve recently seen these two software on HN that maybe of some help:<p>deepfake for voice: <a href="https:&#x2F;&#x2F;github.com&#x2F;CorentinJ&#x2F;Real-Time-Voice-Cloning" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;CorentinJ&#x2F;Real-Time-Voice-Cloning</a><p>Reproducing emotional voices: <a href="https:&#x2F;&#x2F;www.sonantic.io&#x2F;" rel="nofollow">https:&#x2F;&#x2F;www.sonantic.io&#x2F;</a>

You are trying to do a beautiful thing. I don&#x27;t have a knowledge of this subject, but I really wish you good luck on this project.

Is this a time sensitive procedure? I think I’m stating the obvious - (maybe not) - but this is not something you should just wing a few weeks before, nor is it something you should try to figure out on your own without <i>thoroughly</i> discussing with your wife. “Surprise honey, I deepfaked your voice!” is not something most people would appreciate.

Nobody has mentioned VocalID and voice surrogacy [1] yet. This organization might be able to recreate her voice from historic samples for speech-to-text<p>[1] <a href="https:&#x2F;&#x2F;www.ted.com&#x2F;talks&#x2F;rupal_patel_synthetic_voices_as_unique_as_fingerprints" rel="nofollow">https:&#x2F;&#x2F;www.ted.com&#x2F;talks&#x2F;rupal_patel_synthetic_voices_as_un...</a>

All sorts of feels here. I had a positive outcome from exploratory throat surgery that had a chance of obliterating my voice. Prepping the way you are doing is amazing. Please balance it with time well-spent with your wife, being present in the moment. Sounds trite and yet takes focus to not just concentrate in the possible negative future outcome.

Here&#x27;s a story from the San Francisco Chronicle on saving Stephen Hawking&#x27;s voice: <a href="https:&#x2F;&#x2F;www.sfchronicle.com&#x2F;bayarea&#x2F;article&#x2F;The-Silicon-Valley-quest-to-preserve-Stephen-12759775.php" rel="nofollow">https:&#x2F;&#x2F;www.sfchronicle.com&#x2F;bayarea&#x2F;article&#x2F;The-Silicon-Vall...</a>

You might look at what Jamie Dupree has done.<p><a href="https:&#x2F;&#x2F;www.cnn.com&#x2F;2018&#x2F;06&#x2F;15&#x2F;health&#x2F;dystonia-jamie-dupree-radio-no-voice&#x2F;index.html" rel="nofollow">https:&#x2F;&#x2F;www.cnn.com&#x2F;2018&#x2F;06&#x2F;15&#x2F;health&#x2F;dystonia-jamie-dupree-...</a><p>He uses a text-to-speech system that sounds more-or-less like him.

This is very much a short term solution if they are unable to talk immediately after surgery, for up to a few days. My wife used a small portable whiteboard and magic marker to write messages on in the same situation. It worked really well. Even with our 2 year old, it helped her to understand something unusual was going on.

I’m sorry that the both of you have to deal with this. I’ve read many of the replies here and I’m surprised there isn’t already a self-service website that does this. Pay some money, record some text, and boom here’s your voice. Something like this should exist. Someone should build this.

I don&#x27;t really have anything to add to all the helpful comments under your thread. Do the preparation as much as you can, as long as your wife also wants this.<p>You said there is a small chance, so I really wish you and your wife the best of luck that she and her voice will be fine after the surgery.

If she has time get here to read the most common english words. Then parse the text and play the audio for the known words and use traditional speech synthesis for the outliers. It will not be perfect but you can then possibly train an AI to pronounce the outliers.

I would also think of various phrases that need a lot emotion applied. e.g. for sensitive situations like someone&#x27;s death, or for positive feedback like a wedding or a birthday or a thank you<p>Maybe also if she has a favourite book or a favourite quote, get those recorded too.<p>Back it all up!

If you don’t have any children (yet) you should get her to record herself reading some of her favourite children’s books. At the very least she will be able to read along with them. Children’s books are quite sparse, so a page per-track is easy to do.

Just let her read a couple of pieces of texts and record in high-quality (44 KHz).<p>Beyond the techical answer, you may want her to record some nice personal words addressed to your family that you can listen to later.<p>You don&#x27;t need to do anything until the worst case materialises.

There is <a href="https:&#x2F;&#x2F;www.descript.com&#x2F;lyrebird-ai" rel="nofollow">https:&#x2F;&#x2F;www.descript.com&#x2F;lyrebird-ai</a> which is in private beta right now, but looks to serve your needs exactly. Maybe reach out to them?

For recording training audio:<p><a href="https:&#x2F;&#x2F;github.com&#x2F;daanzu&#x2F;speech-training-recorder" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;daanzu&#x2F;speech-training-recorder</a><p>The recorder works with Python 3.6.10. Need to pip install webrtcvad also.

Roger Ebert has some articles about his troubles he encountered that may be worth a read.

Confident as I may be that OPs intentions are good and pure, a quick CTRL-F on the comment threads finds no references to “abuse” or “ethics”, and I propose that synthesis of voice raises issues for which society has few natural defenses.

<a href="https:&#x2F;&#x2F;www.modeltalker.org&#x2F;vrec&#x2F;" rel="nofollow">https:&#x2F;&#x2F;www.modeltalker.org&#x2F;vrec&#x2F;</a> is a project for &quot;voice banking&quot; that might be able to help. It&#x27;s not perfect yet.

Good luck with what you are doing and more importantly, I wish your wife good health.

My first thought was to spend some time together not speaking. See how it goes, so there’s less fear going into it. Maybe take a couples mime class or something! Just making it real and not living in fear is the point.

IDK about the tech, but I would not worry about it right now. You dont need to play with the tech unless the bad unlikely outcome comes to pass.<p>The only tip I have is from a bit of amateur sound editing I did: collect many samples, and beware of big phrases: Like, ask her to say the same thing many times. And ... sometimes ... to ... stop ... at ... each ... word. And ... so ... me ... ti ... mes at each syllable.<p>Otherwise, if you ever need to create a sample that contains a single word&#x2F;syllable, you cant. It is weird how much sound that contains clearly distinguishable syllables for the human ears still is not separable when you go to edit it.<p>Also, you might want to check wordlists by frequency to get a menu of common words, and ipa notation, to ensure you cover a good range of sounds

I recently learned about a startup that is working on this kind of tech: <a href="https:&#x2F;&#x2F;phonetic.ai&#x2F;" rel="nofollow">https:&#x2F;&#x2F;phonetic.ai&#x2F;</a>

Take a look at Lyrebird<p><a href="https:&#x2F;&#x2F;www.descript.com&#x2F;lyrebird-ai?source=lyrebird" rel="nofollow">https:&#x2F;&#x2F;www.descript.com&#x2F;lyrebird-ai?source=lyrebird</a>

Please do keep us posted on the final outcome. We all pray for the surgery to go successful. ( Really appreciate your efforts for preparing for the worst case scenario)

Vocalid.ai has an vocal bank where you can record yourself, and use other people&#x27;s voices as well. It could be a good choice for her to use her own voice

These are the things i am coming always back to ycombinator.com. There are always valuable, intelligent replies here for all kind of issues you might have.

Make sure to not have her read too much. The vocal cords can get inflamed and increase the chance of complications&#x2F;damage.

what dangers are there of someone &#x27;stealing&#x27; your voice to impersonate you later? it seems mostly theoretical right now, but perhaps the more high-profile you are, the bigger the dangers might be, even today? if you had a large body of your voice already recorded (prepped for voice processing systems), is that data high-risk?

Please let us know the good news if they arrive, preferable with Tell HN or something similar.<p>Good luck and best wishes! &lt;3

get a great microphone and have her read her favorite books. Go for books with lots of dialog and emotional content.<p>Later, you can extract all the phonemes you want from it and you will retain the emotional expressiveness of her voice.<p>She should probably sing some songs -- lullabies, rock, etc. Go for emotional diversity.

There&#x27;s an app for that! It&#x27;s called Vocable, it&#x27;s open source and iOS and Android!!!

&gt; I&#x27;d like to prepare, just in case, to have technology to reproduce her voice from keyboard or other input.<p>Is this something that she wants? She&#x27;s got a lot on her plate (emotionally and logistically) to prepare for this surgery, and maybe doesn&#x27;t need a big geek project inflicted upon her just because there&#x27;s a small chance of a bad outcome.

How small of a chance of her losing her ability to speak are you talking about here?

Learn sign language.

Acapela.com has a voice banking service

Please. There is a small chance you said. Everything will be fine. But still carry on your research on the problem since it might help others.

Descript.com has the tech.<p>Reach out to Andrew Mason.

Not bad for as a niche product Idea

Does your wife want you to do this?

Descript.com Andrew Mason

Did you ask her about that? Make sure she is not freaking out of that.