MGM-Omni: Scaling Omni LLMs to Personalized Long-Horizon Speech

Kari Jaaskelainen

30 Sep 2025 — 1 min read

A dual‑track “brain‑mouth” design for unified multimodal understanding and expressive, streaming voice

MGM‑Omni unifies multimodal reasoning with long‑horizon, controllable speech generation in a single Omni LLM. Rather than bolt speech onto a text reasoning core, it uses a dual‑track, token‑based “brain‑mouth” architecture that cleanly decouples multimodal cognition from real‑time speech synthesis—enabling low‑latency streaming and rich cross‑modal interaction.

On the understanding side, a unified training strategy with a dual audio encoder supports robust, long‑form perception across varied acoustic conditions. For generation, a chunk‑based parallel decoding approach narrows the text–speech token‑rate gap, boosting throughput while preserving stability. The model delivers streaming zero‑shot voice cloning that maintains timbre identity over extended spans, alongside natural, context‑aware prosody.

Despite data‑efficient training, MGM‑Omni reportedly outperforms existing open‑source systems on timbre preservation, long‑form audio understanding, and omnimodal comprehension. The result is an end‑to‑end paradigm that brings personalized, controllable speech together with vision‑and‑audio understanding—without resorting to brittle cascades.

Why it matters: voice is the most natural interface, but real‑time, expressive, and personalized speech at scale has been elusive. MGM‑Omni offers a practical architecture for assistants, content creation tools, and accessibility solutions that need long‑horizon voice coherence and tight multimodal grounding.

Paper: arXiv: MGM‑Omni
Register: AiFeta

#SpeechAI #Multimodal #LLM #VoiceCloning #Streaming #AudioUnderstanding #TTS

From f(x) and g(x) to f(g(x)): LLMs Learn New Skills in RL by Composing Old Ones

Evidence that RL teaches genuinely new abilities: compositional skills emerge and transfer across tasks Does RL merely reweight what an LLM already knows—or can it teach genuinely new skills? This paper offers concrete evidence for the latter. Using a controlled, synthetic framework, the authors define “skills” as string transformation

OpenGPT-4o-Image: A Comprehensive Dataset for Advanced Image Generation and Editing

A structured 80k instruction–image corpus spanning 11 domains and 51 subtasks to train unified visual editors Unified models for image generation and editing hit a data ceiling: existing corpora emphasize basic manipulations but miss real‑world complexity. OpenGPT‑4o‑Image tackles this with a hierarchical task taxonomy and automated

Random Policy Valuation is Enough for LLM Reasoning with Verifiable Rewards

ROVER replaces PPO loops with uniform‑policy Q‑values—boosting quality and diversity in math reasoning Popular RLVR methods for LLM reasoning lean on generalized policy iteration (e.g., PPO/GRPO), but suffer instability and diversity collapse. This paper reframes math RLVR as a specialized finite‑horizon MDP with deterministic

CLPO: Curriculum Learning meets Policy Optimization for LLM Reasoning

A dynamic, self‑paced curriculum that restructures problems to match model ability in RLVR Online RL with Verifiable Rewards (RLVR) has boosted LLM reasoning—but most methods treat all problems equally, wasting effort on solved items and flailing on those beyond current capability. CLPO fixes that with a dynamic pedagogy:

Read more

From f(x) and g(x) to f(g(x)): LLMs Learn New Skills in RL by Composing Old Ones

OpenGPT-4o-Image: A Comprehensive Dataset for Advanced Image Generation and Editing

Random Policy Valuation is Enough for LLM Reasoning with Verifiable Rewards

CLPO: Curriculum Learning meets Policy Optimization for LLM Reasoning