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 details the construction of a Voice AI Assistant, integrating speech-to-text (AssemblyAI) and a local LLM (Ollama) for intent detection. The system allows users to perform actions such as creating files and generating code through spoken commands.
The article details the creation of VoiceForge AI, a locally hosted, voice-powered coding assistant and file manager, emphasizing privacy and cost-effectiveness. It explores the architecture, local AI models like faster-whisper, and solutions for engineering challenges such as compound intents and safe execution.
Deepgram's production Speech-to-Text (STT) and Text-to-Speech (TTS) voice models are now natively available on Together AI. This integration provides dedicated model inference for developing real-time voice agents.