A narrator records the same sentence for the fifteenth time. The microphone is fine, the room is fine, and the fatigue is the problem. Across the same city, a merchant is making a product video in six languages and has no budget for six voice actors. Both are customers for a technology that has quietly become mainstream.
AI dubbing (AI配音) is no longer a novelty in short video, audiobooks and online advertising. Mobvoi (出门问问), a Shanghai company listed in Hong Kong as 2438, is one of the firms selling it. Its 2025 annual report gives an unusually clear picture of what this business looks like once the excitement wears off.
Key takeaways
- Mobvoi reported 2025 revenue of ¥323.0 million (≈ US$45M / HK$352M), down 17.24 percent, and a net loss of ¥68.9 million (≈ US$9.6M / HK$75M), narrowed by 90.5 percent.
- AI hardware contributed ¥174.2 million (≈ US$24M / HK$190M), up 14.4 percent; AI software solutions accounted for roughly ¥149 million (≈ US$21M / HK$162M).
- Its products include the Chinese dubbing tool Moyin (魔音工坊) and an overseas version, DupDub, which the company says supports more than 37 languages.
- The value proposition is workflow: script, voice, subtitles and editing in one place, not a better robot voice in isolation.
- The commercial question is no longer whether people will use synthetic speech, but who gets paid for it.
The product is a workflow, not a voice
The obvious way to describe AI dubbing is text-to-speech (TTS): type words, hear them spoken. That description misses why creators actually pay. A working tool has to handle pacing, pronunciation of names, multiple characters in a dialogue, subtitles that line up with the audio, and a license that permits commercial use.
Mobvoi's Moyin product is built around that bundle. Company and industry materials describe a large library of voices, voice cloning (声音克隆) from a short sample, multi-character reading for audio dramas, and export of subtitle files. The overseas product, DupDub, is aimed at merchants making product videos in languages they do not speak. The company says it uses its own speech models and a voice engine called MeetVoice Pro.
For a solo creator, this collapses a chain of purchases. Instead of hiring a narrator, buying a subscription to a transcription service and paying for a commercial-music license, one tool produces the spoken track and the captions. The quality may not match a trained actor, but most marketing clips do not need a trained actor. They need a clean read delivered by tomorrow.
What the financials reveal
The Hong Kong exchange filing for 2025 shows revenue of ¥323.0 million (≈ US$45M / HK$352M), a decline of 17.24 percent from the previous year. The net loss narrowed to ¥68.9 million (≈ US$9.6M / HK$75M), a 90.5 percent improvement, and the adjusted net loss narrowed by 72.6 percent. Gross margin was about 50 percent, and the company had roughly 150 employees at year-end.
Those numbers tell a modest story. AI voice is a real business, but it is not a gold rush for this particular company. Hardware remained the larger revenue line, and software solutions — the segment that includes generative-AI products — was somewhat smaller. The company has described a shift toward an "AI agent plus hardware" strategy and a reorganization around AI-native workflows.
This is worth saying plainly because AI coverage often jumps from a demo to a billion-dollar forecast. The reality for most generative-media companies is slower: a useful product, a paying user base, and a long struggle to grow revenue faster than inference and distribution costs. A narrowing loss is a meaningful achievement in that context, but it is not the same as scale.
The deeper change: whose voice is it
The most interesting consequence of cheap dubbing is not cost. It is the loosening link between a voice and a person. A creator can now license a synthetic narrator, clone a voice with a short sample, or generate a character that does not exist. For small businesses making localized ads, that is liberating. For the people whose voices are sampled, it raises practical questions about consent and payment.
Those questions are not theoretical. Professional narrators and actors have watched synthetic voices enter exactly the low-budget work that used to pay their bills. A company selling a voice library has an incentive to make licensing simple; the existence of a license does not settle whether the source performer was fairly treated. Readers should treat "commercial license included" as a starting point, not the end of the inquiry.
There is also a quality trap. Synthetic speech is convincing enough for a product video and unconvincing in a long documentary, where listeners notice flat emotion and odd emphasis. Used well, it removes drudgery. Used carelessly, it makes every video sound identical, which is its own kind of brand damage.
Why this matters beyond China
Mobvoi is a Chinese company, but the pattern is global. Text-to-speech, voice conversion and automatic dubbing are becoming default features in video editors and e-commerce tools everywhere. The competitive frontier has moved from "can a machine speak?" to "can it speak the right way, in the right language, with a clear right to use the voice?"
For non-Chinese readers, the company's results are a useful counterweight to the idea that generative media is only a Silicon Valley story. A Shanghai firm with ¥323.0 million (≈ US$45M / HK$352M) in revenue and a global-facing dubbing product is participating in the same market as larger Western labs. The difference is scale, not the nature of the problem.
Honest limitations
This is a business-and-trend deep read. Financial figures come from Mobvoi's published Hong Kong annual-results announcement and annual report as summarized by Chinese financial media. Product descriptions and the number of supported languages are company or industry materials, not independent tests; no user counts are used because I could not confirm them against authoritative sources. The article makes no claim about the relative naturalness of Mobvoi's voices versus competitors', and it does not assess any court case, licensing dispute or regulatory rule about synthetic speech in any jurisdiction.
What readers can do now
- If you make videos, try synthetic narration on a draft before hiring a narrator, and listen on a phone speaker — that is where most audiences hear it.
- Check the license attached to any voice you use, especially for paid ads, and keep a record of which voice generated which file.
- If you are a narrator or voice actor, read the terms before uploading a sample for cloning; a short recording can have a long life.
