CASTIUM Release v1.2.2 Patch Notes

Release Date: September 27, 2026
Installer Package: CastiumSetup-v1.2.2.msi

What's New in v1.2.2

🎙️ Free Offline Local Neural Voices (Kokoro-82M & Piper TTS)

• Kokoro-82M Studio Quality Neural Voices (100% Local & Free):
  • Embedded local AI voice synthesis using a weight-quantized 82M parameter model running completely on CPU via high-speed ONNX runtime under open-source Apache-2.0 license with zero cloud dependencies or subscriptions.
  • Delivers warm, highly expressive human-like radio voiceover with realistic breath cadence, natural emotion, and fluid diction.
  • Packaged studio voices: Bella, Sarah, Nicole, Sky, Michael, Adam, George.
  • High-performance 44.1 kHz stereo PCM resampler with seamless 2-bus audio output.
  • 1-Click setup and downloader in Application Settings (⚙️ Settings → 🧠 AI & LLM).
• Piper Offline Neural Engine:
  • Ultra-fast, lightweight neural TTS running in sub-100ms response times on CPU under the MIT license.
  • HD voice models (~35 MB each): Lessac, Bryce, Amy, Ryan.
  • Dynamic DJ Surfacing: Downloaded voices immediately surface on the DJ persona bar and selection dropdown with live download progress tracking.

🎵 Universal Live Track & Song Recognition

• Zero Metadata Loss Across All Audio Pipelines:
  • Unified track detection (GetLiveOnAirTrackInfo()) cascading across:
    1. Castium Local Player (LocalPlayer.CurrentTrack)
    2. Dual Decks (Deck A & Deck B playing or loaded track)
    3. Windows Media Session Capture (Spotify, Apple Music, Deezer, YouTube, web browsers, and desktop media players via WASAPI loopback capture).
• Automatic External Track Transitions:
  • Song transitions in Spotify, web browsers, and media players now automatically notify the AI DJ service (OnTrackStarted), triggering autonomous track intros and banter seamlessly regardless of music source.
• Strict Track & Artist Prompt Enforcement:
  • AI Brain prompt and fallback smart templates now strictly bind the active track and artist ("Task: Announce the music on air. You MUST explicitly name the track '{activeSong}' and artist '{activeArtist}'. Keep it under 25 words.").

🤖 Autonomous AI DJ Banter & Random Scheduling Engine

• Master Autonomous On/Off Switch:
  • Prominent [ ● BANTER ON ] (green) / [ ○ BANTER OFF ] (gray) toggle on the AI Station Co-Host card.
  • Toggles and schedules are now persisted across application restarts in %USERPROFILE%\.castium\ai_dj\brain_config.json.
• Flexible Broadcast Scheduling Modes:
  • 🎲 Random (2–4 Songs): Dynamically randomizes track rotation (introducing a track every 2 to 4 songs) for an authentic, unpredictable radio broadcast feel.
  • ⏱️ Random Timer (8–15 Min): Independent background timer that triggers spontaneous station banter every 8 to 15 minutes regardless of track boundaries.
  • ⚡ Smart Hybrid (2–4 Songs or 12 Min): Blends random track-based intros with an automatic 12-minute silence safeguard so your station never goes too long without lively DJ presence.
  • Fixed Intervals: Every 1, 2, 3, or 5 songs.
• 1-Click Manual Trigger ([ 🧠 Banter Now ]):
  • Instant manual banter button in the quick liners bar for immediate live commentary on demand.

🧠 AI DJ Content Brain Enhancements & Model Support

• Full Support for Heavy Local Models (Ollama 6B–8B):
  • Increased internal generation and HTTP request timeouts from 8s to 45s/60s to comfortably support large local Ollama models (such as qwen3-vl:latest, llama3.1:8b, and vision-language models) without premature timeouts or template fallback.
• Real-Time UI Loading Feedback:
  • Live [ ⏳ Generating... ] visual button state with real-time status notices while local LLMs process tokens.
  • Generated banter text automatically populates the Custom Script Pad so DJs can review the script in real time.
• Markdown Script Sanitization:
  • Automatically strips markdown bold/italic asterisks (**Queen** → Queen) and parenthetical stage cues before sending text to speech synthesis engines.
• Immediate Auto-Save:
  • Changing intelligence providers (Ollama, Cloud API, Smart Presets) or broadcast vibes (Top 40, Chill, Rock, Punchy) immediately auto-saves to your user profile.

🎚️ Analog Studio Vocal DSP & Auto-Ducking

• Broadcast Channel Strip:
  • Voice synthesis routes through an analog studio microphone processor: Proximity Bass Warmth (+2.5 dB at 150 Hz), High-Frequency Air (+1.8 dB at 4.2 kHz), Studio Acoustic Room Space (18ms booth reflection to eliminate dry synthetic deadness), and Soft Vocal Leveling.
• Smooth Analog-Grade Music Ducking:
  • Music beds smoothly duck by -12 dB with analog-style exponential curves whenever the AI DJ speaks and bloom back up upon completion.