OpenAI 发布 GPT-Live-1 系列语音模型，实现更自然的实时对话与全双工交互

OpenAI 今日发布了名为 GPT-Live-1 和 GPT-Live-1 mini 的新对话模型，声称它们听起来更自然，并能更好地处理轮流发言。这些是全双工模型，意味着它们可以同时说话和聆听，允许用户自然地打断对话，并支持实时翻译等功能。

该公司还将默认用 GPT-Live-1 mini 替换 ChatGPT 中现有的高级语音模式。付费用户将能够使用更大的 GPT-Live-1 模型。先前的模型结合了语音转文本模型用于转录语音、大语言模型生成回复以及文本转语音模型给出最终答案。

在一次媒体简报中，该公司表示，新模型解决了在用户说话时打断他们以及回答不够智能等问题。OpenAI 的新模型将在继续对话的同时，将查询发送给其最新的文本模型（如 GPT-5.5），以进行搜索、推理或实现智能体能力。

OpenAI 还展示了该模型可以长时间保持沉默并吸收对话上下文，直到被呼唤。另外，由于新的语音模式可以访问较新的 GPT 模型，它还能以视觉格式呈现某些信息。其他创业公司如 Monogram，该公司已从 DST 和 Lux Capital 获得 4000 万美元种子轮融资，也在探索视觉化回复，以使助手更具互动性。

公司称，ChatGPT 中的新语音模式旨在进行更长时间的对话。在简报会上，ChatGPT 语音产品负责人 Atty Eleti 表示，他在散步时曾与该语音功能进行过 30 到 40 分钟的对话。

OpenAI 认为，对于复杂工作而言，语音可能成为计算的主要交互界面。有报道称，它可能在今年推出一款具备 AI 功能的耳机。然而，该公司没有提供任何有关硬件产品的信息。

Eleti 表示：“随着时间的推移，我们认为这还将释放使用语音作为计算主要界面并管理日益复杂的长时间智能体工作的能力。我们看到人们通过 Codex 和 ChatGPT 实现的那些令人惊叹的用例，我们相信语音可以成为未来各种工作的交互界面。”

过去几年，OpenAI 一直在增强基于语音的功能，以使 ChatGPT 的语音模式听起来更自然。该公司表示，超过 1.5 亿人使用语音和听写等功能与 ChatGPT 交谈。

竞争对手也在尝试让助手更具表现力。

无论是苹果还是亚马逊，都更新了它们的助手，使其更善于对话并更好地处理上下文。创业公司如由 Oculus 联合创始人 Brendan Iribe 和 Ankit Kumar 创立的Sesame，也推出了能在后台完成任务的同时进行更自然对话的 AI 助手。

OpenAI 正朝着同一方向前进，目标是让用户能更长时间免提与助手交谈。尽管声称新语音模式听起来更自然，但该公司强调，其目标并非将其打造成 AI 伴侣。它指出，新模型内置了安全措施，能为青少年提供适龄的回复，并在对话涉及自残等话题时提供相关资源。

新的语音模式仍有待改进。在演示过程中，当公司展示其印地语实时翻译功能时，助手带有浓重的美国口音，讲出的印地语听起来不自然，略带学究气。该公司表示新模式针对“大多数使用口语的语言”进行了优化，但没有具体说明是哪些语言。

OpenAI today released new conversational models, called GPT-Live-1 and GPT-Live-1 mini, claiming that they sound more natural and can handle turn-taking better. These are full-duplex models, meaning they can speak and listen at the same time, allowing users to interrupt naturally and enabling features like live translation.

The company is also replacing its current Advanced Voice Mode in ChatGPT with GPT-Live-1 mini by default. Users of paid tiers will be able to access the larger GPT-Live-1 model. The previous model combined a speech-to-text model to transcribe speech, a large language model to generate responses, and a text-to-speech model to deliver the final answer.

The company said in a press briefing that the new models solve issues like interrupting users while they’re talking and not having enough intelligence to answer questions. OpenAI’s new models will send the query to its latest text models like GPT-5.5 for search, reasoning, or agentic capabilities while continuing the conversation.

OpenAI also showed that the model can stay silent for a long time and absorb the context of the conversation until it’s called upon. Plus, as the new voice mode has access to newer GPT models, it can also present some information in a visual format. Other startups like Monogram, which raised $40 million in seed funding from DST and Lux Capital, are also leaning into visual responses to make assistants more interactive.

The company said the new voice mode in ChatGPT is designed to have longer conversations. During the briefing, ChatGPT Voice’s product lead, Atty Eleti, said he has had 30- to 40-minute-long conversations with the voice feature during walks.

OpenAI thinks that voice could be the primary interface to computing for complex work. Reports have suggested that it could launch a pair of earbuds with AI capabilities this year. However, it didn’t provide any information on hardware products.

“Over time, we think this will also unlock the ability to use voice as a kind of primary interface to computing, and to manage increasingly complex long-running agentic work. The kind of amazing use cases that we see people using Codex and ChatGPT to accomplish, we think voice can be the future interface to all kinds of work,” Eleti said.

OpenAI has worked on bolstering voice-based features over the past few years to make ChatGPT’s voice mode sound more natural. The company said that more than 150 million people talk to ChatGPT using features like Voice and Dictation.

Rivals are also attempting to make assistants more expressive.

Both Apple and Amazon have updated their assistants to be more conversational with better context handling. Startups like Sesame, founded by Oculus co-founder Brendan Iribe and Ankit Kumar, also launched AI assistants with more natural conversation while completing tasks in the background.

OpenAI is moving in the same direction, aiming to let users talk to its assistant hands-free for a longer time. Despite its claim that the new voice mode sounds more natural, the company emphasized that it’s not aiming to make this an AI companion. It noted that the new models have safeguards built in to give age-appropriate responses to teens and provide resources if the conversation turns to topics like self-harm.

The new voice mode still needs work. During the demo, when the company showed its live translation feature in Hindi, the assistant had a heavy American accent and spoke in Hindi that was unnatural sounding and had slightly bookish tone. The company said the new mode is optimized for “most spoken languages” but didn’t specify which ones.