SUB/WAVE

Product Introduction

1. Definition: SUB/WAVE is a self-hosted, AI-powered internet radio station software stack. It is a technical solution that integrates a Large Language Model (LLM) as an autonomous DJ, a Liquidsoap-based audio mixer, and an Icecast server to create a single, synchronized audio stream from a personal music library.
2. Core Value Proposition: It exists to recreate the authentic, shared experience of traditional radio broadcasting using modern, private AI technology. Its primary value is delivering a single Icecast stream where every listener hears the same broadcast simultaneously, curated and announced in real-time by an AI DJ, all while maintaining complete data privacy and user control through self-hosting.

Main Features

1. Agentic AI DJ System: The core intelligence is a language model that autonomously manages the broadcast. It analyzes context (time, weather, recently played tracks, listener requests) to select the next song from a connected Subsonic-compatible music server (like Navidrome). It then generates and delivers live voice announcements (station IDs, song intros, time checks) using a configurable Text-to-Speech (TTS) engine. This happens in real-time without pre-recorded segments.
2. Unified Broadcast Architecture: The system is built on a deterministic, single-stream model. All audio processing (crossfading, voice ducking, jingle insertion) is handled by Liquidsoap, and the final mixed output is served as a single MP3 (and optional Opus/AAC/FLAC) stream via Icecast. This ensures every listener is synchronized, replicating the "live radio" feel, as opposed to personalized, asynchronous streaming.
3. Modular & Swappable AI Stack: A key technical feature is the complete separation of the "brain" (LLM) and the "voice" (TTS). The LLM layer is provider-agnostic, supporting local models (via Ollama, llama.cpp), major cloud APIs (OpenAI, Anthropic, Google), and aggregators (OpenRouter). Similarly, TTS can be local (Piper, Kokoro) or cloud-based (ElevenLabs, OpenAI). This allows operators to choose their balance of cost, privacy, and voice quality, and even implement zero-shot voice cloning for custom DJ personas.
4. Comprehensive Operator Console: The admin dashboard provides granular control over the station. Features include a 24/7 programming grid for scheduling shows with specific personas and music moods, a library of autonomous "skills" (e.g., weather reports, news digests) for the DJ, a real-time booth log, manual track queuing, a playlist generator, and system diagnostics. It centralizes all configuration for personas, voices, jingles, and the music mood taxonomy.

Problems Solved

1. Pain Point: The loss of shared, communal listening experiences in the age of personalized, algorithmic playlists. It addresses the desire for a "lean-back" music discovery medium that isn't controlled by the listener's immediate clicks, fostering a sense of community around a single broadcast.
2. Target Audience: Music enthusiasts and homelab operators who want to broadcast their personal libraries; podcasters and community builders looking for a always-on audio channel; developers and tinkerers interested in applied AI/LLM agents, audio streaming, and self-hosted software.
3. Use Cases: Creating a private "family radio" station for a household; running a themed internet radio station for a niche community or fan club; setting up ambient background music for a coffee shop or workspace; using it as a novel interface and discovery engine for a large personal music collection; experimenting with AI persona design and voice synthesis in a practical audio project.

Unique Advantages

1. Differentiation: Unlike music servers (Plex, Navidrome) that serve individual streams or playlist apps, SUB/WAVE is a true broadcaster. Unlike internet radio automation software (e.g., RadioDJ, Azuracast), it uses a dynamic, context-aware LLM for curation and live voice instead of static playlists and pre-recorded liners. Unlike cloud AI DJ services, it is self-hosted, private, and uses the user's own music library without licensing fees.
2. Key Innovation: Its agentic architecture treats the LLM as a real-time decision-making agent within a closed-loop broadcast system. The integration of modular, swappable AI components (LLM + TTS) into a stable, file-based audio pipeline (Liquidsoap/Icecast) is a unique technical approach. The request system using natural language processing to search a local library and its focus on synchronized listening are also distinctive features in the self-hosted music space.

Frequently Asked Questions (FAQ)

1. How does SUB/WAVE work with my music library? SUB/WAVE does not store music itself. It connects to a Subsonic-compatible music server like Navidrome, Airsonic, or Gonic via its API. The AI DJ queries this server to search, browse, and stream tracks directly from your existing, self-hosted music collection.
2. Do I need an OpenAI API key to run the AI DJ? No. The software is designed to run fully locally using Ollama and local TTS engines like Piper, requiring no internet API calls or fees. Cloud LLM and TTS services are optional for users who prefer them.
3. Can listeners skip songs or control the broadcast? No, to preserve the authentic radio experience. Listeners join a single, synchronized Icecast stream. The only listener interaction is the natural language request system, where a song is matched by the AI and added to the central queue for everyone to hear.
4. What are the technical requirements to self-host SUB/WAVE? You need a machine (like a home server, NAS, or VPS) capable of running Docker containers. It will host the SUB/WAVE stack (Node.js controller, Liquidsoap, Icecast), your chosen LLM (e.g., Ollama), TTS engine, and your Subsonic-compatible music server. Sufficient CPU/RAM for audio encoding and AI inference is required.
5. Is SUB/WAVE free to use? Yes. SUB/WAVE is open-source software released under the MIT license. There are no subscription fees or tiered plans. You are responsible for the cost of your own infrastructure and optional paid cloud AI APIs if you choose to use them.