CLAP Audio Encoder Tested on ESC-50 Environmental Sounds via ONNX Runtime

·1 views

A developer exported the CLAP audio encoder from the laion/clap-htsat-unfused model to ONNX format and evaluated it on the ESC-50 dataset, which contains 2,000 five-second environmental sound recordings across 50 classes. The experiment used a fixed one-nearest-neighbor approach, selecting two clips per class for a total of 100 clips, with no classifier training involved. Embeddings were compared using cosine similarity to assess whether recordings from the same class clustered together and whether sounds aligned with broader category groupings. The pipeline ran on an Apple Silicon CPU using PyTorch 2.8.0 and ONNX Runtime, producing normalized 512-dimensional embeddings per audio clip. The full code and reproducible environment are publicly available in the kiarina/labs repository on GitHub.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

React Query Simplifies Async Data Fetching, Caching, and State Management in React

React Query is an open-source library designed to handle asynchronous data fetching, caching, and state management in React applications. It replaces manual data-fetching boilerplate with a single hook, ensuring consistent patterns across development teams. The library's intelligent caching system serves stored data instantly while silently re-fetching updates in the background, improving perceived app speed. React Query draws a clear separation between server state and client state, reducing the complexity typically associated with tools like Redux. It also ships with built-in error handling and automatic retries with exponential backoff, minimizing the impact of transient network failures on users.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

How Trivy Catches Security Flaws in Dockerfiles and Terraform Before Deployment

Static Application Security Testing (SAST) tools can scan Infrastructure as Code (IaC) configuration files — such as Dockerfiles and Terraform scripts — to detect misconfigurations before any infrastructure is provisioned. In a walkthrough by developers Fabrizio Salvador and Elias Perez Peralta, Trivy, an open-source security scanner, was used to audit a sample containerized application managed via Terraform on AWS. A single command flagged three vulnerabilities: a container running as root, an unencrypted S3 bucket, and an SSH port exposed to the public internet. The team then remediated each issue by enforcing a non-root user, enabling AES-256 server-side encryption, and restricting SSH access to an internal VPN range. A follow-up Trivy scan confirmed all failures were resolved, demonstrating how SAST can be integrated into CI pipelines to enforce security standards early in development.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Solo Dev Ships 284k Lines of TypeScript in Two Months, Says Distribution Is the Real Challenge

A solo developer built Astroniq, an AI-powered Vedic astrology platform, committing over 832 times across five frontend apps, four backend services, and fifteen shared packages in just two months. The project featured advanced computational tools including divisional charts, agentic AI chat, and birth-time refinement using real ephemeris data. The developer documented several critical infrastructure pitfalls encountered along the way, including a Docker deployment sequence that could cause full outages and a disk-space issue where build cache silently accumulated until it crashed Redis. These technical lessons were shared to illustrate that with modern AI tooling, writing code is no longer the primary bottleneck for a startup. The developer concludes that shipping a product is now achievable solo at speed, but other challenges — such as distribution — remain the harder, less-discussed obstacle.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Meronq Ships Desktop MVP Shell in Sprint 9 Using Tauri 2 Framework

The Meronq development team has closed Sprint 9, delivering the first desktop shell built on top of the project's existing packages. The native application, scaffolded with Tauri 2, wraps a web UI and allows users to connect a project folder, view handshake and memory statistics, and manage an MCP server. The desktop shell integrates with existing Meronq packages including server-core, memory, and mcp-server, in line with the project's ADR-0008 architectural decision. Users can initialize a new project or connect an existing repository, with the MCP server spawning automatically using the configured project path. Upcoming work includes native build verification for Windows and macOS, a memory search interface, and a cloud workspace API spike.

0 comments Read more at DEV Community