Show HN: Speechos – Benchmark 25 speech AI models locally, no cloud needed (github.com)

Speechos has launched a local benchmarking platform for a variety of speech AI models, allowing users to evaluate speech-to-text (STT), text-to-speech (TTS), and emotion recognition capabilities without relying on cloud services. Users can easily record or upload audio, seamlessly switch between different STT and TTS engines, and receive instant results, all while maintaining data privacy—no audio ever leaves the local machine. This tool supports multiple engines such as Whisper and Vosk for STT and various TTS systems, facilitating comparisons through detailed analyses of accuracy, emotional response, and speaker identification. The significance of Speechos lies in its capacity to empower developers and researchers to assess model performance directly on their hardware, enhancing decision-making for tailored applications. Key technical features include a Python backend and a web interface powered by Node.js, supporting quick model selection and task execution via a REST API. Notably, users can benchmark models dynamically, analyze audio features, and receive automatic emotion scores, making it a versatile resource for experimentation in the rapidly evolving AI/ML landscape.