Boson AI launches Higgs Audio v3

// 1h agoMODEL RELEASE

Boson AI launches Higgs Audio v3

Boson AI has launched Higgs Audio v3, a 4B parameter text-to-speech model built on a Qwen3-4B backbone and optimized for real-time conversational streaming and zero-shot cloning across 100+ languages. The model supports inline style tags for prosody and emotion control, and integrates with the newly released SGLang-Omni inference framework for low-latency deployment.

// ANALYSIS

Conversational voice AI is transitioning from slow, turn-based systems to low-latency, continuous streaming, making Higgs Audio v3 and SGLang-Omni crucial for realistic real-time agents.

* Integrating with SGLang-Omni allows Higgs Audio v3 to begin speech synthesis before a sentence finishes, resolving a critical latency bottleneck.

* Granular inline tags enable developers to dynamically control speaker emotions and sound effects, making applications feel far more interactive.

* Releasing weights under a non-commercial research license drives community adoption while retaining commercial monetization for Boson AI.

// TAGS

ttsspeechvoice-cloningboson-aispeech-synthesissglang-omniaudio-model

DISCOVERED

1h ago

2026-06-07

PUBLISHED

1h ago

2026-06-07

RELEVANCE

8/ 10

AUTHOR

AI Search

// KEEP READING

More AI developer news from the feed

EXPLORE FULL FEED

NEWS44m ago

OpenAI Codex CLI criticized over exit prompts

Developer Kun Chen highlights a design flaw in the OpenAI Codex CLI harness, where typing "exit" does not terminate the session locally but is instead sent to the LLM as a prompt. This lack of local interception results in an unnecessary API call and inference cycle, wasting energy and user tokens.

RESEARCH1h ago

Stability AI releases Stable-Layers, a reinforcement learning framework that trains image layer decomposition models without paired supervision data using Flow-GRPO and VLM feedback.

Stability AI has introduced Stable-Layers, a reinforcement learning framework designed to train image layer decomposition models without requiring paired training datasets. Traditionally, splitting a flat image into editable, multi-layer components required intensive human annotation. Stable-Layers bypasses this by adapting Group Relative Policy Optimization (GRPO) for flow-matching models (Flow-GRPO) to optimize image decomposition. The training is guided by a Vision-Language Model (VLM) serving as a judge, using a structured scoring and grid-based calibration pipeline to provide high-quality reward signals. This approach significantly reduces color bleed and blank layer artifacts, producing cleaner semantic separation.

LAUNCH1h ago

NVIDIA announces the Arm-based RTX Spark superchip combining Grace CPU and Blackwell RTX GPU to run high-performance AI agents locally on Windows PCs.

NVIDIA has introduced the RTX Spark, an Arm-based system-on-chip (SoC) combining a Grace CPU and Blackwell-architecture RTX GPU with up to 128GB of unified memory. Designed to run high-performance AI agents locally on Windows PCs, the superchip delivers up to 1 petaflop of FP4 AI performance, enabling consumers to run large language models of up to 120 billion parameters without relying on cloud infrastructure.

Boson AI launches Higgs Audio v3