LocalAI-io/LocalVQE-demo

1---2title: LocalVQE Demo3emoji: 🎙️4colorFrom: blue5colorTo: green6sdk: gradio7sdk_version: 5.49.18app_file: app.py9python_version: "3.12"10pinned: false11license: apache-2.012models:13  - LocalAI-io/LocalVQE14---15 16# LocalVQE Demo17 18Interactive demo for [LocalAI-io/LocalVQE](https://huggingface.co/LocalAI-io/LocalVQE) —19compact open-source models that clean up a microphone signal on CPU20in a single causal pass. The selector offers the joint models21(acoustic echo cancellation + noise suppression + dereverberation:22v1.3 at 4.8 M parameters, v1.2 at 1.3 M, plus older releases for23A/B), and **v1.4-AEC** (203 K parameters) which removes *only* the24echo — near-end speech, background noise, and room acoustics are25kept intact by design.26 27Pick one of the bundled examples or upload your own `(mic,28far-end-reference)` 16 kHz wav pair. For a pure noise-suppression29test, upload silence (or nothing) as the reference. With v1.4-AEC30selected, noise-only clips should come back nearly unchanged —31that's the intended behaviour.32 33Built with Gradio. Inference runs through the released **GGML C++34engine** (the same `liblocalvqe.so` production users build from the35repo) loading the published `.gguf` files — what you hear is the36deployed artifact, not a Python re-implementation. CPU-only; a3710-second clip takes well under a second on the Space's default tier.38 39Code: [github.com/localai-org/LocalVQE](https://github.com/localai-org/LocalVQE) ·40Training: [github.com/localai-org/LocalVQE-train](https://github.com/localai-org/LocalVQE-train).41