This content features a live demo of LocalVQE, a tiny AI model with approximately 1 million parameters, designed to cancel echo and noise in real-time. It showcases efficient audio processing capabilities.
Llama.cpp (llama-server) now officially supports Speech-to-Text (STT) capabilities, integrating the Gemma-4 E2A and E4A models. This update brings advanced audio processing to the popular open-source AI inference platform.
This article describes the creation of the first public audio dataset for coffee roasting first crack detection, addressing a significant gap in available resources. The dataset, comprising 973 annotated 10-second segments, was meticulously built from scratch and led to a model achieving 100% precision thanks to careful data splitting and loss weighting.
This content explores audio spectrogram representations for processing with Convolutional Neural Networks. It focuses on how these visual representations of sound can be efficiently utilized by CNNs for various tasks.
This technical report details the architecture, methodologies, and performance metrics of the VibeVoice system. It provides an in-depth analysis of its underlying technology and implementation.
Meta introduced SAM Audio, the first unified multimodal model for audio separation. This development marks a significant step in AI for sound processing.
This content explores the phenomenon of hallucination in the Whisper model, explaining why transcripts might loop the same phrase. It details the causes behind this behavior when the model processes periods of silence.