Sources: the vendor's own pages (linked above) and creator videos at the linked timestamps. EU status is information, not legal advice; see how we check.
MiDashengLM-Gen is an audio generation framework from Xiaomi that produces complete 16 kHz mono audio scenes in a single pass, blending speech, music, sound effects and environmental ambience. It uses a Qwen3-1.7B LLM backbone (fully fine-tuned) with the DashengTokenizer and a 16-layer flow-matching DiT, generating 768-dimensional audio latents autoregressively with per-token conditional flow matching. A learned stop head allows variable-length output.
Input is a structured multi-view caption using special tokens: overall caption, ASR transcript, speaker characteristics, sound effects, music and environment, with an unknown token for absent elements. Inference uses a 10-step Euler ODE solver with CFG scale 2.0. The project reports speech intelligibility approaching dedicated TTS systems on Seed-TTS, support for 9 languages and emotion control.
The project page offers audio demos across seven categories and cites an arXiv paper (2608.11804). The page states Apache License 2.0. A YouTube creator reported a model size under 12 GB and a GitHub release, but no repository or Hugging Face URL was present in the provided material, so weight availability is not confirmed here.
Watch the moment: the exact second top creators show it. Videos load only when you press play, via youtube-nocookie.com.
Generates complete audio scenes including speech, music, sound effects and ambient noise in one generation, with emotion and multilingual support. Under 12 GB, open-sourced on GitHub.
Closest neighbours by meaning, not just by category.
Model · Xiaomi
Xiaomi's MiMo family of open-weight language, multimodal, audio and speech models, with a hosted API and web demo.
Research · MiniMax
Open-weight multimodal model from MiniMax whose video pipeline can be run audio-only to generate music and sound effects.
Model · Meta
Meta research model that generates audio directly in raw waveform space, without a VAE, for video-to-audio synthesis.
Model · Multimodal Art Projection (M-A-P)
Open-weights music generator that plans an editable ABC-style score, then renders full songs with vocals and accompaniment.
Model · ACE Studio
Open-source local music generation model with a 4B-parameter DiT decoder, extending ACE-Step 1.5 with higher audio quality.
Model · Xiaomi
Xiaomi's MiMo-V2.6 model series, with a 1T-parameter Pro and a 311B Flash variant, plus a Mimo Desktop harness.
