From WeChat’s AI Podcast Trial to Google, ByteDance and Xiaohongshu: Can AI Podcasts Capture the Emerging AIGC Blue Ocean?

With large language models achieving remarkable progress in dialogue text generation and high‑fidelity voice synthesis maturing, the podcast format—traditionally creator‑heavy—is being reshaped by AI. WeChat recently launched a "Quick News" feature that includes AI‑generated podcast episodes, explicitly labeled as "AI generated," marking a gray‑scale test of the technology.

Achieving the "human feel" in AI podcasts relies on modern neural‑network TTS. Unlike traditional concatenative synthesis, contemporary TTS models capture prosody, timbre, speed, emotion, and style through deep learning, and further benefit from adversarial training, large‑model‑based voice modeling, and multimodal conditioning, making synthetic speech increasingly indistinguishable from real human voices.

Examples of cutting‑edge models include Microsoft’s VibeVoice‑1.5B, released in August, which uses continuous speech tokenization and a next‑generation diffusion tokenizer to handle long‑sequence audio efficiently. Face‑to‑face, the collaborative VoxCPM model (0.5B parameters) from Mian Intelligent and Tsinghua Shenzhen International Graduate School adopts an end‑to‑end diffusion‑autoregressive architecture to generate continuous speech representations directly from text, improving naturalness and timbre similarity. Bilibili’s IndexTTS‑2 introduces a novel, universal duration‑control method for autoregressive TTS, becoming the first model that supports precise timing control.

Current AI podcast ecosystem can be divided into two camps. Large‑tech firms have accelerated attention to the field: Google’s NoteBookLM offers concise audio summaries in over 50 languages, ByteDance’s Doubao leverages the Volcano Engine to generate podcasts end‑to‑end with high naturalness in Chinese, and Xiaohongshu’s audio team released the conversational model FireRedTTS‑2 (arXiv paper titled "FireRedTTS‑2: Towards Long Conversational Speech Generation for Podcast and Chatbot").

Start‑ups demonstrate diverse innovation. Laifu Radio claims to be a "personal AI radio" with fully AI‑generated shows; ChatPods, founded by Zhang Yueguang, provides a personal "AI podcast agent" that creates voice‑summarized content and recommends customized podcasts; Huxe, founded by former NotebookLM members, offers DeepCasts for on‑demand, personalized AI podcasts.

At the 2025 "Made on YouTube" event, YouTube CEO Neal Mohan announced AI tools, including an audio‑video generation system tailored for podcasters to create video clips from podcast audio, illustrating how AI is permeating the entire production pipeline.

From the creator’s perspective, AI podcasts lower content‑creation barriers, assisting script writing, editing, recommendation, and distribution, enabling individuals and small teams to produce high‑quality programs quickly. From the listener’s side, AI‑driven recommendation and voice‑assistant integration promise more efficient and immersive consumption.

According to the 2024 Podcast Industry Report, 45.9% of surveyed users purchased paid podcasts in the past year, and 63.6% are receptive to podcast advertising, indicating growing commercial value. The article concludes that AI podcasts are in a flourishing, multi‑player stage, with technical advances and market demand suggesting a promising future for the AIGC‑driven audio market.