A person finishes a 300-page web novel on the subway, not by reading but by listening — to a synthetic voice that costs the platform almost nothing to produce. Multiply that by hundreds of millions, and you get a market flip that happened in about a year.
AI-narrated audiobooks (有声书) have become one of the clearest consumer wins for generative speech in China, and the business consequences are already visible.
How the flip happened
The trigger was 番茄畅听 (Fanqie Changting), ByteDance's free, ad-supported audiobook app. Its trick is straightforward: take the vast library of free web novels from the Fanqie Novel (番茄小说) ecosystem and convert them into spoken audio with AI text-to-speech, at bulk scale. Users pay nothing; they sit through ads or do in-app tasks instead.
That model worked. By June 2025, Fanqie Changting's monthly active users reached about 129.6 million, according to industry data cited in mainstream coverage, making it the largest single audiobook app by MAU (129.6M). Ximalaya (喜马拉雅) — the paid-subscription leader that had dominated long-form audio for a decade — still reported a larger all-scenario ecosystem of more than 300 million monthly active users and about 144 minutes of daily listening, but its paid model suddenly looked exposed.
The champion got acquired
On May 18, 2026, Tencent Music Entertainment Group (腾讯音乐) completed its acquisition of Ximalaya, announced a year earlier. The deal was worth up to US$1.26 billion in cash plus up to roughly 175 million Tencent Music A-class shares — an implied total around US$2.6 billion (≈ HK$20 billion). China's market regulator had conditionally approved it on May 12, attaching commitments that Tencent not raise prices, not lock in exclusive licensing, and not bundle audio into car systems to squeeze rivals.
The logic was strategic. Tencent Music's music business is short-form and near saturation; long audio — novels, podcasts, courses — fills multi-hour commuting and chore windows. Ximalaya's library and the AI pipeline that is collapsing production cost made it the obvious piece to buy rather than build.
Why AI narration is the engine
The cost math is the story. Industry estimates put human narration of a 100,000-character book at around ¥20,000 (≈ US$2,780 / HK$21,700) with a 15-day-plus turnaround, while an AI full pipeline runs in the tens-to-hundreds-of-yuan range, at per-minute costs of roughly ¥0.5–5 (a few US cents to about US$0.70, or HK$5.4) versus ¥300–800 (≈ US$42–111 / HK$326–868) for human voice. Ximalaya has said AI-narrated audiobooks accounted for about 43% of its audiobook revenue in early 2026, and that AI has replaced more than 60% of mid-to-low-end standardized narration.
Behind the voices, the sector is scaling fast. China's audiobook (有声阅读) user base is estimated at around 580 million in 2026, and the market booked about ¥13.4 billion (≈ US$1.86B / HK$14.5B) in revenue in 2025, according to the Audio-and-Visual Publishing Association of China, with those figures cross-reported by industry press. Podcast listenership tops 150 million. Generative speech is the lever turning cheap text into listenable hours.
The same engine is spreading beyond books. Education providers use AI narration to localize courses into dozens of languages at a fraction of studio cost, and e-commerce sellers clone a single spokesperson's voice to pitch the same product in Mandarin, English and Arabic. The common thread is that speech — once a craft requiring a human in a booth — has become a configurable output, priced by the minute. That lowers the floor for small creators and squeezes the middle of the voice-acting market at the same time.
What readers can do now
- If you consume audiobooks, try a free AI-narrated title and a human-narrated one back-to-back; the gap has narrowed for straightforward narration but widens for emotional or character-heavy performance.
- If you produce audio, pilot AI narration for your back-catalogue and standard content where human cost blocks release, and reserve human voice for premium, character-driven work.
- If you are a voice actor or narrator, expect mid-to-low-end standardization work to keep shrinking; the durable demand is for direction, performance and rights-cleared character voices that AI cannot easily replace.
Honest limitations
This article covers China's long-audio market and generative speech; it does not address the separate copyright questions around training TTS voices on performers' work, nor the quality ceiling of current synthesis for nuanced performance. Some figures (AI revenue share, replacement rate, per-minute costs) come from industry analyses rather than audited company filings, and should be read as directional. The acquisition value is taken from Tencent Music's disclosure and mainstream reporting; share-price-based implied totals move with the market.
Key takeaways
- ByteDance's free, AI-voiced Fanqie Changting crossed 129.6 million monthly users by mid-2025 to become the largest single audiobook app by MAU, even as Ximalaya retained a larger all-scenario ecosystem.
- Tencent Music closed its acquisition of Ximalaya on May 18, 2026 (up to US$1.26 billion cash plus shares), betting long audio offsets music's saturation.
- AI narration has cut per-minute production cost from hundreds of yuan to single digits, and now drives a large and fast-growing share of China's ~580-million-user audiobook market.