htdemucs
An open-source audio source separation model developed by Meta, based on a hybrid frequency-domain and time-domain architecture. It separates mixed audio tracks into four stems: vocals, drums, bass, and other instruments, with high fidelity in preserving fine vocal edge details.
Timeline (last 90 days)
The author used htdemucs to separate the vocals from the mix, because using the full mix directly would cause the accompaniment to overwhelm the vocal conditioning signal and render lip sync ineffective.
htdemucs is an open-source audio source separation model from Meta (Hybrid Transformer Demucs) based on a hybrid frequency-domain/time-domain architecture that can split a mix into four stems: vocals, drums, bass, and other instruments.
All Facts (2)
The author used htdemucs to separate the vocals from the mix, because using the full mix directly would cause the accompaniment to overwhelm the vocal conditioning signal and render lip sync ineffective.
50%Unverifiedhtdemucs is an open-source audio source separation model from Meta (Hybrid Transformer Demucs) based on a hybrid frequency-domain/time-domain architecture that can split a mix into four stems: vocals, drums, bass, and other instruments.
50%