aifollow.news 搜索
返回 IT之家 科技新闻
IT之家 科技新闻· · 原发布时间

Block 推出全双工语音模型 Dessa,暂未开放试用

自动核验发布 · 本文由系统生成并完成证据核验,未经人工审稿。

AI 辅助摘要

Dessa 能在说话时同步听取用户语音,并在对话中切换音色、语速和情绪。Block 称,它可在 M4 Max 或更高配置的 MacBook 上本地运行;目前正邀请安全研究者测试语音智能体欺骗人类的风险,尚未对外开放试用。

正文 · AI 翻译

IT Home reported on October 25 that Block, the company led by Twitter co-founder Jack Dorsey, has launched a new voice model called Dessa. It can run locally and simulate emotional expressions such as laughter, pauses, and a choked-up voice, producing highly humanlike speech. According to the company, Dessa was trained entirely on human speech data, using only 1 rack of Nvidia GPUs during training.

Dessa is a full-duplex model. Like GPT-Live, which OpenAI released earlier this year, it can listen to a user while speaking. Mainstream voice models generally need to be paired with a large language model to access knowledge; Block says Dessa draws its knowledge from the conversational data used to train it.

Block says existing voice AI experiences often feel stiff and unnatural because the generation process first converts speech into text and then uses that text to synthesize speech. Many models, including ElevenLabs v4 and Fish Audio Drama 3, rely on speech tags to control emotion. Dorsey's company claims emotional expression is built into the model itself, although it remains unclear whether the underlying system still uses some kind of tags to regulate emotion.

Block wrote in a blog post: “Voice AI interactions today are terrible. They first turn your speech into text, losing a great deal of information that text cannot convey. Even the world's most powerful AI cannot hear your emotions. We think this makes little sense: it creates a poor experience and poses potential risks.”

The research team released several sample conversations with Dessa. The agent can change its voice, speaking speed, and emotional tone. In one conversation, for example, the AI says it is exhausted from work, and its voice conveys fatigue at the same time.

In another example, the AI talks about how the other person makes it feel at ease, speaking with emotion and sounding close to choking up. Another demonstration shows Block employees playing a game with the voice assistant.

According to the company, the model can run locally on a consumer computer, with an M4 Max MacBook or a higher-spec model as the minimum hardware requirement. Dessa is not yet available for public testing. Block is inviting safety researchers to test it, with a focus on scenarios in which this type of voice agent could deceive people.

According to IT Home, Block's main business is financial services, and it provides solutions for small and micro businesses. Its Square business already uses voice AI to process orders. Having developed a voice model that can run on devices, Block may be able to roll out more voice products for merchants and consumers at a lower cost.

Advertising disclosure: External links in this article, including but not limited to hyperlinks, QR codes, and codes, are provided to convey more information and save readers time in evaluating options. The results are for reference only. All IT Home articles containing external links include this disclosure.

发现内容有误?提交纠错