For a deaf student in a Chinese university lecture hall, the gap between "I understand" and "I'm lost" often comes down to a single missing feature: captions. Most classrooms, meetings, and livestreams assume everyone in the room can hear.
That blind assumption is why a seven-year-old Chinese program keeps quietly scaling. iFlytek's (科大讯飞) "听见AI的声音" (Hear the Voice of AI) campaign converts spoken language into on-screen text in real time — and hands it to deaf and hard-of-hearing users for free.
Key takeaways
- Built on iFlytek's 讯飞听见 (iFlyrec) real-time speech-to-text engine, the program has donated over 200 million minutes of free transcription to nearly 100,000 certified deaf users as of May 2026.
- It launched in 2019 alongside the China Association of the Deaf (中国聋人协会) and the China Disabled Persons' Performing Art Troupe (中国残疾人艺术团).
- In 2025 it powered the first accessible broadcast of the CCTV Spring Festival Gala (央视春晚), captioning nearly five hours of live programming.
- iFlytek also helped draft a national captioning standard, GB/T 44882-2024, which took effect on June 1, 2025.
A country where 27.8 million can't hear
China is home to roughly 27.8 million people with hearing disability (听力残疾), per China Disabled Persons' Federation (中国残疾人联合会) figures — the single largest category of disability in the country. Yet professional sign-language interpreters number fewer than 10,000. For most deaf people the everyday friction isn't one dramatic barrier but a thousand small ones: a hospital intake form, a job interview, a parent-teacher meeting, a livestream they can't follow.
Speech-to-text changes the equation. Instead of needing a human interpreter physically present, a phone or laptop can render a speaker's words as scrolling text within seconds.
What the program actually does
At its core, "听见AI的声音" runs on 讯飞听见, iFlytek's real-time automatic speech recognition product. A deaf user opens the app, certifies their hearing status, and unlocks:
- Real-time transcription of meetings, classes, interviews, and livestreams, with a "no-clear-screen" mode so text doesn't scroll away mid-sentence.
- A text-input box so the user can type replies that the other party reads — turning a one-way broadcast into a two-way conversation.
- 讯飞同传 (iFlytek Simultaneous Translation), free for over a year, supporting Chinese plus 14 other languages — useful for study, travel, and cross-border communication.
- 讯飞译制 (iFlytek Production), which adds AI subtitles and multilingual dubbing to videos, letting deaf creators localize their own content.
The technology is not magic. It is large-model (大模型) automatic speech recognition tuned for accuracy in noisy, fast-paced, accented real-world speech — the kind human interpreters handle by instinct but machines still find hard.
From a charity line item to a standard
The scale is what makes this more than a PR campaign. According to iFlytek, cumulative free transcription passed 200 million minutes by May 2026, serving nearly 100,000 certified users. Those figures are self-reported, but they line up with the program's inclusion in the Ministry of Industry and Information Technology's (工业和信息化部) "2025 Digital Accessibility Products and Services Typical Cases" list — one of just 15 internet-application demonstration cases.
The program also reached a cultural milestone in early 2025. The CCTV Spring Festival Gala (央视春晚), watched by hundreds of millions, used iFlytek's technology for its first accessible broadcast, generating real-time captions across nearly five hours of live content through an AI-plus-human workflow. For the first time, many deaf viewers could follow the gala's jokes and lyrics as they happened, not the next day.
Standards work followed the deployment. iFlytek participated in drafting GB/T 44882-2024, "Information technology — closed captioning" (信息技术 可关闭字幕), released in November 2024 and effective June 1, 2025. The standard lets closed captions be encoded as an independent stream alongside audio and video — meaning subtitles can be switched on or off like sound, a building block for accessibility across Chinese streaming and broadcast.
In 2024, iFlytek and the China Disabled Persons' Federation (中国残联) also set up a joint general-AI assistive-technology laboratory (通用人工智能助残联合实验室) to push from software into physical rehabilitation aids.
According to Li An, chief scientist at BrainNet (脑机网), "real-time captioning is the lowest-friction entry point for assistive AI, because it needs no new hardware — just a phone and a model that hears well."
Honest limitations
This piece relies on figures published by iFlytek and China Disabled Persons' Federation-affiliated outlets; independent third-party audits of the 200-million-minute and 100,000-user claims are not publicly available, so treat them as company-reported. The service requires users to upload proof of hearing disability to unlock free tiers, which some may find intrusive. Real-time transcription still struggles with heavy accents, overlapping speakers, and domain jargon, and the free simultaneous-translation tier caps hours per month. Coverage is strongest in Mandarin and major cities; rural and minority-language support is thinner. I have not tested the app firsthand.
What readers can do now
- If you work with deaf or hard-of-hearing colleagues in China, point them to 讯飞听见's hearing-mode certification — the free transcription tier is real and substantial.
- Product teams: read GB/T 44882-2024 and treat closed captions as a first-class, toggleable stream, not an afterthought.
- Outside China, the lesson is architectural: real-time captioning is the cheapest, highest-leverage accessibility feature you can ship — start with meetings and live video.