Speech2Speech – browser-based STT and TTS for interacting with local LLMs (github.com)

Speech2Speech is a new web application that allows users to seamlessly convert speech to text and back again using local language models, all processed within the browser without any data transfer to external servers. It leverages the Moonshine speech recognition model for transcribing spoken English into text, sends that transcription to a locally hosted language model API for analysis, and then employs the Kokoro text-to-speech engine to vocalize the AI's response. This user-friendly tool features a dark mode for comfort and requires a local server like llama-server to operate. This development is significant for the AI/ML community as it emphasizes privacy by keeping all data processing local, reducing potential data exposure risks associated with using cloud services. It incorporates advanced technologies such as Hugging Face's Transformers.js for machine learning capabilities, the Web Audio API for handling audio, and modern JavaScript features for enhanced performance. By ensuring compatibility with local or remote servers and providing a straightforward setup process, Speech2Speech enables users to interact with AI in a more personal and secure way, showcasing the potential for browser-based applications to enhance human-computer interaction.