Bao Yu Says Human Narration Still Beats AI for Children’s Audiobooks

What happened

Bao Yu says AI-generated audiobooks are usable but still fall short of professional human narration, and pronunciation errors with Chinese characters that have multiple readings could mislead children. For those making an audiobook with AI, he suggests having an LLM check pronunciation first, then using an Agent to operate TTS with readings, pauses, and emotion marked. He says paid options are generally better than open-source ones, but does not recommend a provider; he uses Gemini TTS for convenience, not because he considers it the best.

The story so far

Bao Yu (Dotey) shared advice for making AI audiobooks, saying the results are usable but still fall short of professional human narration. He warned that errors with Chinese characters that have multiple pronunciations could mislead children. For people making their own AI audiobooks, he recommended first having a large language model check pronunciation, then using an agent to operate text-to-speech with pinyin annotations for characters with multiple readings, pauses, and emotion marked before generation. He said paid options are generally better than open-source ones but did not recommend a particular provider. He uses Gemini TTS for convenience, not because he considers it the best.

Keep reading

Trending now

1. 1Mistral and NVIDIA Announce Industrial Engineering AI Collaboration and Open Benchmarks
2. 2Anthropic Opens Claude Duet Research Preview With Waitlist for US Pro and Max Users
3. 3OpenAI Models and Codex Workflows Coming to Google Cloud, With Reports Differing on Availability
4. 4Mistral Releases Voxtral 3 Speech Models, Voice Agents API and LiveKit-Based SDK