Voice Interaction

Voice interaction is presented as a still-open opportunity for AI-native products. The host argues in 高手怎么用 AI？普通人怎么学 AI？投资人如何投 AI？｜对谈课代表立正 that typing is not the most comfortable human interface, especially for Chinese input, and that voice can carry richer information than text. 我遇到了第一个真正想买的陪伴机器人！｜对话世博：越伴动力创始人【公路播客】 adds the opposite design lesson for Companion Robots: Xiaoban uses non-human sounds instead of ordinary speech so users can feel intent without expecting a generic talking assistant. 把7位黑客松选手请进播客｜冠军、怪才和48小时不眠的野心家 adds the assistive branch through Kenan Voice Changer, where AI repairs unclear speech to help communication.

71. 编程的内燃机时代 adds the translation branch. The hosts discuss AI translation earbuds and compare the experience to a “Babel fish”, connecting voice interfaces to AI Translation and cross-language communication.

171: 【AI季报 26Q2】从 coding 到 RSI，强者愈强的未来？ adds the full-duplex model branch through Thinking Machines Lab and Interaction Model. The source argues that voice is not just another multimodal input: useful spoken AI should listen and speak at the same time, handle interruptions, see user actions, and feel closer to a phone call than a push-to-talk assistant.

Source Notes

• The episode connects voice to early WeChat usage patterns and voice messages.
• It also mentions interactive audio experiences, including a game controlled by spoken pitch.
• Voice is framed as both an interface and a content format.
• The Xiaoban source shows voice-like expression can be intentionally non-verbal and paired with posture, gaze, and gesture.
• The Kenan Voice Changer source shows voice interaction can be an accessibility layer, not only a convenience interface or entertainment format.
• The 内核恐慌 source shows voice interaction can also be a cross-language layer through real-time translation.
• The LateTalk source shows voice interaction moving toward full-duplex, interruptible, real-time assistant behavior through Interaction Model.

Connections

• WeChat — platform context for voice and social interaction.
• Everything Agent — agents may need more natural interaction channels.
• Video Models — adjacent content-generation direction.
• Xiaoban, Robot Liveliness, and Emotional Interaction Models — companion-robot case where voice is one channel among embodied signals.
• Assistive AI, Kenan Voice Changer, and Li Pengcheng — accessibility case where voice repair supports communication.
• AI Translation and Immersive Translate — cross-language branch added by Neihe Konghuang.
• Thinking Machines Lab and Interaction Model — full-duplex voice model branch added by LateTalk.