Image: coindesk.com
The Story
Dessa sounds human in demonstrations; whether it can pass as a particular person has not been independently tested.
Dessa’s natural interruptions, laughter and emotional timing could make a conversation feel real, but that is not the same as reproducing one person’s identity. The missing evidence is decisive: no public benchmark tests Dessa against named voices, and Block has not disclosed model size or performance results.
- Public accounts describe natural conversational timing and emotional cues, but they do not document Dessa impersonating a specific individual.
- Block has not disclosed model size or benchmarks, and the preview is limited to safety researchers rather than general users.
- Independent reporting and research show that voice cloning can enable scams, but safeguards and test conditions strongly affect how convincing a clone is.
Deeper Insight
Dessa brings live impersonation to the MacBook, where interruptions, laughter and timing can make a fake caller feel present.
A MacBook that listens, interrupts, laughs and answers in real time gives voice scams a live performance, not just a recording. Dessa’s full-duplex design lets automated callers handle the messy cues that make conversation feel human, moving impersonation toward ordinary devices. When voices can perform presence, identity has to rely on secrets beyond speech.
- Existing local systems have cloned a user’s voice from short reference recordings, showing that on-device voice imitation is already practical in some settings.
- Voice-cloning research identifies fraud, defamation and erosion of trust as central risks when systems reproduce specific people’s voices at scale.
- Dessa’s full-duplex design could make deception more persuasive by handling interruptions, acknowledgements, laughter and changing tone in real time.
Sources · 10
- 1 Journals & institutions
- 2 News reporting
- 3 Forum, video, personal
- 4 Other / unclassified
What do these mean?
- Journals & institutions. Published in a peer-reviewed journal, or from a university, government body, court or other institution.
- News reporting. Journalism: newspapers, magazines and news sites.
- Forum, video, personal. Forums, social posts, video, blogs and personal accounts.
- Other / unclassified. A site we have not classified yet. This says nothing good or bad about it.
Featured · Analysis · techcrunch.com
Block’s new voice model sounds eerily human
News reporting
Block's new voice model sounds eerily human | TechCrunch Image Credits: screenshot via Revolution.Social (opens in a new window) # Block’s new voice model sounds eerily human Ivan Mehta 9:00 AM PDT · October 24, 2026 Jack Dorsey’s Block released a new voice model out of the company’s research lab that runs locally and sounds strangely human with its expressions like laughs, pauses, and cries. The
Read sourceAnalysis · particle.news
Particle: Block Unveils Dessa, a Speech‑Only AI That Listens While It Talks
Other
Particle: Block Unveils Dessa, a Speech‑Only AI That Listens While It Talks # Block Unveils Dessa, a Speech‑Only AI That Listens While It Talks Block is limiting early access to select safety researchers to study the impersonation and trust risks of a human‑like, full‑duplex voice model. ## Overview
Read sourceAnalysis · hottarita.substack.com
ローカルTTSを試す【第6回】Qwen3-TTSをMacで動かして声を比べた - by hottarita ほったりた
Forum or personal account
ローカルTTSを試す【第6回】Qwen3-TTSをMacで動かして声を比べた - by hottarita ほったりた # ローカルTTSを試す【第6回】Qwen3-TTSをMacで動かして声を比べた hottarita ほったりた 7月 30, 2026 こんばんは、ほったりたです。 6月に「ローカルTTSを試す」という連載をやっていて、 第5回で VoxCPM に自分の声をクローンさせるところまで書きました。 そのあと僕は、その仕組みでラジオを作りはじめました。 つまり VoxCPM に落ち着いてしまったのです なので、今日は、別のローカルTTSを1本入れて、 同じセリフを同じ参照音声で読
Read source