Speech to Text for Deaf and Hard of Hearing People

Speech to text for deaf and hard of hearing people usually means live captions: the words of other people, shown as text as they speak. Many phones and computers now include a built-in option, usually part of the operating system, so there is no separate app to install. Automatic captions have not yet matched trained human captioners in research. For a lecture, a medical visit or any conversation where a missed word matters, the standard remains human captioning, called CART. The law also requires reasonable access to captioning and other aids at work, at school and in health care, where it applies.

Live captions turn the speech around you into text in real time

Live captioning shows other people's speech on your screen as they talk. The need is large. About 15 percent of American adults, or 37.5 million people, report some trouble hearing ¹. That figure is self-reported, and it rises with age. In the 2014 National Health Interview Survey, 5.5 percent of adults aged 18 to 39 reported trouble hearing ². The share was 19.0 percent for adults aged 40 to 69 ³ and 43.2 percent for adults aged 70 and older ⁴.

Worldwide, the World Health Organization estimates that 430 million people need rehabilitation for disabling hearing loss ⁵. The American figure counts self-reported trouble hearing, and the WHO figure counts disabling hearing loss, so the two should not be added together.

Live captions are harder than a transcript you read later. Each word has to appear within seconds of being spoken, so both the delay and the accuracy matter. The sections below compare the options by setting: built-in captions, group conversation, captioned phone calls, human CART, smart glasses and the legal rights that apply.

Major phones and computers now include built-in live captions

The built-in options cover video, podcasts and face-to-face talk with one or two people, as the table shows. They are the first thing to try, because they come with the operating system and need no separate app.

On a Mac, the captions can come from the computer's audio or from its microphone. Apple's guide says you can switch between them, "such as during an in-person conversation" ¹⁰. Apple's iPhone guide also warns that accuracy "may vary" and that the feature should not be relied on in high-risk or emergency situations ⁶.

On Windows 11, live captions can run without an internet connection, and they can include microphone audio so that in-person conversations are easier to follow ⁸. Microsoft's page describes what the feature does, not how accurate it is.

On Android, Google introduced Live Caption with the Pixel 4 in 2019. At launch, it captioned videos and spoken audio with one tap, and it left out phone and video calls ⁹.

Yap, which we make, is built for dictating your own speech on Mac and Windows, not for captioning other people, so it is not one of the options in this section. For captioning someone else's speech, use the built-in features above, a captioned phone service, or a human captioner.

Conversation captioning works best one to one in a quiet room

Built-in captions work best when one person speaks clearly, close to the microphone, in a quiet room. Group conversation is harder for the software and for the reader. Researchers who asked deaf, hard of hearing and hearing people about automatic captions in group conversation found that the main obstacles were text accuracy and lag time ¹¹. The same authors note that recognition accuracy also depends on ambient noise and microphone quality ¹².

Accurate text still leaves a gap. The study's introduction cites earlier research on real-time text. Citing that research, it reports that deaf and hard of hearing people received 50 to 80 percent of the information from accurate captions, against 84 to 95 percent for hearing peers ¹³. A caption shows the words but not who is speaking or where to look, so the reader has to work that out while following the conversation.

These steps help in practice:

1. Sit where you can see the speaker and the screen at the same time, so you do not lose one while reading the other.
2. Ask people to take turns. One speaker at a time gives the software and the reader the best chance.
3. Use the microphone setting on a Mac for in-person talk, so the captions come from the room rather than from a video (see the table above).
4. Put the device close to the speaker and away from fans, traffic, music or a television.

Step 4 follows from the noise finding above. The other steps are practical advice, not findings from the studies.

Captioned telephone services caption phone calls, and a federal fund pays for the captions

Android's Live Caption did not caption calls when Google launched it, so phone calls need a different service. A captioned telephone shows the other caller's words as text while you listen. Federal rules call this an internet protocol (IP) captioned telephone, or IP CTS ¹⁴. For background on what the service is and how to start, see what a captioned telephone is.

Eligibility is limited. Federal rules say that anyone other than registered users with hearing loss may not use IP captioned telephones with captions turned on ¹⁴. Providers must print that notice on newly distributed equipment and in their advertising and education materials ¹⁴.

The required consumer notice says there is a cost for each minute of captions, paid from a federally administered fund ¹⁵. The notice says who pays for the captions, not what your phone bill will show.

Each provider sets its own enrollment steps. Check its registration and eligibility questions before you sign up, and compare how each one handles your phone and your hearing.

Human CART captioning is still the standard for lectures, medical visits and legal settings

When a missed word matters, use a human captioner. The Department of Justice's ADA guidance calls this real-time captioning, also known as computer-assisted real-time transcription, or CART. A transcriber types what is said into a computer, and the words are projected onto a screen ¹⁶. The CART captioning entry explains the service in more detail.

Automatic captions miss more words than trained captioners

Trained human captioners still lead in the research. The authors of a 2024 study report that existing approaches for improving automatic recognition still fail to match trained CART captioners ¹⁷.

Automatic accuracy also varies from one system to another. In a 2024 study, eleven speech recognition services transcribed higher education lecture recordings, and accuracy "ranges widely between vendors and for the individual audio samples" ¹⁸. The study also found significantly lower quality for streaming recognition, the mode used for live events ¹⁹.

Word error rate correlates weakly with what deaf readers find usable

The standard measure of automatic accuracy is word error rate ²¹, which measures the discrepancy between machine and human transcripts. A 2017 study with 30 deaf and hard of hearing participants tested the authors' new metric, called ACE, and found that its correlation with the participants' ratings of caption usability was 0.74 (Spearman rho) ²². Word error rate correlated only 0.11 with the same ratings ²³.

The practical lesson is that a vendor's headline accuracy figure is a weak guide to whether captions will work for you. Test a captioning app on the speakers, rooms and microphones you actually use before you rely on it. For more on how accuracy is measured, see the guide to speech recognition accuracy.

Medical visits, lectures and legal settings each have a different route

Medical visits. The ADA requires covered entities to provide auxiliary aids and services when needed to communicate effectively with people who have communication disabilities ²⁴. The guidance lists real-time captioning among those aids. Ask for CART before a consultation where accuracy matters, rather than relying on automatic captions on your phone.

Lectures. Ask the school's disability office for CART or a human captioner for each class, and say when a class has technical vocabulary. The rights that support this are in the last major section of this guide, on disability law, and the guide to speech to text for students with disabilities covers the planning steps.

Court and legal settings. If you need captioning in court or another official proceeding, ask the clerk or the court's disability coordinator early, because a human captioner may need to be booked in advance.

The built-in options are a good match for casual conversation and video. CART is the right choice when you need a full, accurate record: lectures, meetings where decisions are made, and medical visits.

Caption glasses put text in your line of sight, with real limits

Caption glasses show speech in your line of sight, so you do not look away from the speaker. The evidence so far comes from research prototypes and small studies. In the sources we checked, we did not find independent accuracy measurements for commercial caption glasses, so the right test is a trial in your own settings. For the short answer, see do speech to text glasses work for deaf people.

With glasses, you read captions at a glance, so delay is a key test. A University of Toronto report on AR captioning sets a benchmark: captions should never lag more than 5 seconds, and 80 percent of captions should appear within 3 seconds ²⁵. That report is a student project, not a peer-reviewed standard, and its own latency test had 5 participants ²⁵.

The strongest study we found is a randomized field experiment. At a school for the deaf, AR captioning could enhance learning performance ²⁶. The same paper found the benefit appeared only in some contexts: where the captions gave nuanced information, and for students with strong visual and auditory processing ²⁷. That is a useful result, but it is not a promise that glasses work equally well for every deaf or hard of hearing person.

Glasses inherit the limits of the speech recognition behind them. The streaming-quality finding and the noise finding above may apply to glasses as much as to a phone, so test the glasses in the room where you will use them.

Disability law gives deaf people rights to captioning at work and school

In the United States, the Americans with Disabilities Act (ADA) requires reasonable access to captioning and other aids in work, school and public services. The rules differ by setting, and they are general information rather than legal advice. A specialist, a disability advocate or the organization's accommodation office can tell you how they apply to your case.

Employers must accommodate you unless it causes undue hardship

The ADA covers employers with 15 or more employees ²⁸. A covered employer must make a reasonable accommodation for a known disability, unless the accommodation would impose an undue hardship ²⁸. For options at work, read the guide to dictation as a workplace accommodation.

Schools must consider what each student needs

Under ADA Title II, a public college must give primary consideration to a disabled student's request when it chooses an auxiliary aid or service ²⁹. Primary consideration is not the same as getting the exact tool you request, because the college may choose an equally effective alternative.

For children in public schools, the federal IDEA regulations require the IEP team to consider whether a child needs assistive technology devices and services ³⁰. An Individualized Education Program is the written plan that records those decisions.

In England, Scotland and Wales, Access to Work can pay for assistive software

Outside the United States, the rules differ. In England, Scotland and Wales, GOV.UK says an Access to Work grant can help pay for specialist equipment and assistive software, and that the grant does not have to be paid back ³¹.

Test any caption option in your own setting before you rely on it

Accuracy claims describe test audio, and your setting may be different. Before you depend on a captioning option, run a short trial with the people and places you actually need:

1. Use the speakers you talk with most. Include any accent, name or technical vocabulary you meet every day. Accuracy varies by speaker, so one good demonstration is not enough.
2. Use the real room and device. Test from the seat you will sit in, with the microphone where it will be and the background noise you expect.
3. Look for meaning errors, not only word errors. A wrong name, number or "not" can change what was said. In one 2017 study, word error rate correlated only 0.11 with deaf and hard of hearing users' ratings of caption usability ²³.
4. Check the delay. Watch whether the captions keep up with the conversation, because a caption that arrives late may be read after the moment has passed.
5. Ask what the vendor's figure measured. Find out whether it covers live or recorded audio, since live streaming recognition measured lower in one study ¹⁹.
6. Book a human captioner for the high-stakes moments. A lecture, a medical appointment or a legal hearing should not depend on a trial you ran on a quiet afternoon.

Choose the option that matches the setting

Start with the built-in captions on the device you already use, and test them in the rooms where you will need them. Use a captioned telephone for calls, and ask for human CART when a missed word would cost you something. Treat smart glasses as promising but unproven for your setting until you have tried them. And know your rights: the ADA covers employers and public services, and IDEA requires schools to consider assistive technology.

Related guides in this series: speech to text for dyslexia and dictation for blind and visually impaired people. For captions on video and meetings, see the guide to captions and subtitles and the question on whether automatic captions are ADA compliant. For whether automatic captions are accurate enough for you, see the question on AI captioning accuracy. The full set of guides on accessibility is in the accessibility and health overview.