Mirello raises 59.1 billion won in investment... Adds drum sounds to muted videos using AI

Source: TokenPost Original Title: Mirelo raises $59.1 million in funding… Using AI to add drum sounds to muted videos Original Link: Germany-based audio generation AI startup Mirelo(Mirelo) has recently attracted attention by raising $41 million(approximately 5.91 billion KRW) in a seed funding round. Index Ventures and Andreessen Horowitz co-led the investment, with Atlantic Ventures and TriplePoint Capital also providing funding.

Mirelo is focused on developing audio generation features tailored to videos, and its flagship model, ‘SFX 1.5,’ successfully recognizes drum scenes in muted videos and generates sound with high precision, mimicking real drum performances. According to independent external judges, this model significantly outperforms existing commercial audio generation models. It is particularly competitive in reducing unwanted noise and accurately syncing audio with scenes in fast-moving videos.

This model is available through the ‘Mirelo Studio’ app, which is accessible via API and for general users. Users can generate various versions of audio through text prompts and select the best results for further editing. The platform is designed to make AI audio creation accessible even to those without technical expertise.

Building on this investment, Mirelo has begun developing next-generation models beyond SFX 1.5. The new algorithms are expected to enhance consistency and natural connectivity between audio segments, with plans to expand into high-quality content creation fields such as movies and video games.

The company is also utilizing high-performance clusters based on NVIDIA’s H100 and H200 GPUs for training its models. These clusters operate through Slurm(Slurm) distributed processing systems. The AI framework adopts Meta’s PyTorch(PyTorch), actively leveraging the Fully Sharded Data Parallel (FSDP) feature optimized for large-scale model distributed training.

Currently, Mirelo is hiring AI researchers specialized in diffusion(diffusion) and autoregressive(autoregressive) architectures. Notably, the autoregressive models include transformer architectures, which are emerging as key technologies not only in language models but also in audio AI applications such as music generation.

With the secured funding, Mirelo plans to strengthen its research team, diversify its product offerings, and accelerate its entry into the global market, aiming to establish a leading position in the sound generation AI industry.