TheMindExpansionNetwork/night-shift-voice-pack

Night Shift Voice Pack — Multi-Character AI Voice Models

Overview

Four unique AI voice characters for the Night Shift autonomous radio show. Each voice is an OmniVoice clone reference pack — just 3-10 seconds of reference audio that can be used with k2-fsa/OmniVoice to generate speech in that voice.

Voice Characters

🎤 MindExpander (Main Host)

• Voice ID: mindexpander
• Role: Main Host
• Source: Jimsky's real voice, fine-tuned OmniVoice checkpoint-1500
• Style: Energetic, confident, late-night talk show host
• Model: TheMindExpansionNetwork/omnivoice-mindexpander-voice

🌙 DJ Nexus (Late Night DJ)

• Voice ID: dj_nexus
• Role: Late Night DJ / Co-Host
• Source: Cloned from Expresso dataset (speaker ex01, male)
• Style: Deep, smooth, laid-back radio broadcaster
• Reference: 2.2s sample — "Why are you beating up my jukebox?"

⚡ Luna (Tech Co-Host)

• Voice ID: luna
• Role: Tech Co-Host
• Source: Cloned from Expresso dataset (speaker ex02, female)
• Style: Energetic, witty, fast-talking AI enthusiast
• Reference: 7.2s sample — "There's an alarm tomorrow at twelve fifteen AM..."

🔮 The Architect (Mysterious Producer)

• Voice ID: architect
• Role: Mysterious Producer
• Source: Cloned from Expresso dataset (speaker ex03, male)
• Style: Deep, resonant, speaks rarely with gravitas
• Reference: 3.8s sample — "No! Doesn't it bother you to sleep around without being in love?"

Usage

Via API

# Generate speech with any voice
curl -X POST https://m1ndb0t-2045--mindforge-omni-api.modal.run/v1/audio/speech \
  -H "Content-Type: application/json" \
  -d '{"text": "Your text here", "voice": "dj_nexus", "guidance_scale": 2.0}'

Available voices: mindexpander, dj_nexus, luna, architect

Via OmniVoice Python API

from omnivoice import OmniVoice

model = OmniVoice.from_pretrained("k2-fsa/OmniVoice")
voice_pack = load_voice_pack("TheMindExpansionNetwork/night-shift-voice-pack")

for voice_id in ["mindexpander", "dj_nexus", "luna", "architect"]:
    audio = model.generate(
        text="Hello, this is a test.",
        ref_audio=voice_pack[voice_id]["ref_audio"],
        ref_text=voice_pack[voice_id]["ref_text"],
        guidance_scale=2.0,
        num_step=32,
    )

Night Shift Show Format

The 4-voice cast engages in a roundtable radio show format:

1. MindExpander hosts and introduces topics
2. DJ Nexus provides smooth transitions and music commentary
3. Luna breaks down the tech/AI aspects
4. The Architect drops in occasionally with wisdom/mystery

License

MIT — Reference audio sourced from Expresso dataset (Meta AI, CC-BY-NC 4.0). MindExpander voice is proprietary (Jimsky's real voice).