VQV Signal

ROBOTS & HARDWARE WATCH TECHNICAL

Haru-4B: a 4B open speech model that listens while it talks, per an X post

According to a post on X by @HuggingModels, Haru-4B is a 4B speech-to-speech model built for full-duplex conversation, meaning it can listen while it talks. The post says it processes audio every 133 ms, works in nine languages and runs in about 12 GB of VRAM. The post is listed under Organization News and links to the original post on X. The provided snippet does not include benchmarks or release terms.

Source: Scobleizer News Pulse (X) · x.com Found via Robert Scoble’s Scobleizer News Pulse · a post on X, not independently verified Published 2026-11-20T05:40:00+00:00 Detected 2026-11-20T19:02:38+00:00
View original source

According to a post on X by @HuggingModels, Haru-4B is a 4B speech-to-speech model built for full-duplex conversation, meaning it can listen while it talks. The post says it processes audio every 133 ms, works in nine languages and runs in about 12 GB of VRAM. The post is listed under Organization News and links to the original post on X. The provided snippet does not include benchmarks or release terms.

AI-assisted summary based on the listed source.

Haru-4B is a 4B speech-to-speech model built for full-duplex conversation, processing audio every 133 ms in nine languages and running in about 12 GB of VRAM. Why it matters: Natural back-and-forth voice agents that run on a single consumer-class GPU. Original post on X Section: Organization News

The post frames the model as a way to build natural back-and-forth voice agents that run on a single consumer-class GPU. Builders and businesses testing voice agents may want to check the model's documentation and independent results before relying on these figures. Watch for the original release details and any third-party testing.

Hardware and robotics watchers may want to track whether this becomes a product, benchmark, or deployment signal.

Signal Strength 92% Technical label WATCH Public Interest 32 Category ROBOTS & HARDWARE Reader Depth TECHNICAL

Signal Strength reflects source quality, relevance, freshness and evidence. Public Interest helps organize discovery; it is not proof of truth.

Public Interest components
Recognizable Entity Score 0 Practical Impact Score 8 Novelty Interest Score 94 Consequence Score 18 Curiosity Score 48 Shareability Score 46

VQV surfaced this signal because it is recent, relevant to AI Voice, connected to Scobleizer News Pulse (X).