Research Demos

Interactive demos for prosody control techniques. Each demo lets you try a different approach to emotion and prosody manipulation.

Current Focus: S2 - Temporal Keyframe Control

These demos explore different ways to control prosody at word-level. Goal: "word 3 = angry, word 7 = calm" with smooth transitions.

Try it

Voxgenesis

Inference script for VoxGenesis latent speaker manifold.

  • Generated by frontend-designer
Open demo
Try it

Voxcpm

VoxCPM Inference Script

  • Generated by frontend-designer
Open demo
Try it

Style Lora

Generate Speech with Style LoRAs and DCFG (Dual Classifier-Free Guidance)

  • Generated by frontend-designer
Open demo
Try it

Steering

Inference with Activation Steering for Training-Free Emotion Control

  • Generated by frontend-designer
Open demo
Try it

Spherical

Inference with Spherical Emotion Vectors (EmoSphere-TTS approach)

  • Generated by frontend-designer
Open demo
Try it

Segment Aware

Inference script for Segment-Aware Conditioning.

  • Generated by frontend-designer
Open demo
Try it

Prosody

Controllable Voice Generation with Prosody Conditioning

  • Generated by frontend-designer
Open demo
Try it

Promode

ProMode Inference Script

  • Generated by frontend-designer
Open demo
Try it

Pitchflow

PitchFlow Inference Script

  • Generated by frontend-designer
Open demo
Try it

Mpe Tts

Inference script for MPE-TTS multi-modal emotion control.

  • Generated by frontend-designer
Open demo
Try it

Mismatch Aware

Generate speech with mismatch-aware CFG emotion control.

  • Generated by frontend-designer
Open demo
Try it

Indextts2

Inference script for IndexTTS2: 8-Dimensional Emotion Vector Control

  • Generated by frontend-designer
Open demo
Try it

Hed

Inference with Hierarchical Emotion Distribution (HED) - V6

  • Generated by frontend-designer
Open demo
Try it

Facodec

Generate Speech with FACodec Prosody Conditioning

  • Generated by frontend-designer
Open demo
Try it

Emovoice

EmoVoice Inference Script

  • Generated by frontend-designer
Open demo
Try it

Emorl

Generate speech with EMORL-TTS: VAD + Local Emphasis Control

  • Generated by frontend-designer
Open demo
Try it

Eme Tts

Generate speech with EME-TTS emphasis-emotion coupling.

  • Generated by frontend-designer
Open demo
Try it

EmoKnob

Direction vector emotion control with intensity slider

  • Emotion selection
  • Intensity 0-2x
  • Real-time preview
Open demo
Try it

EASV Intensity

Emotion-Adaptive Spherical Vectors with smooth interpolation

  • Neutral-to-emotion blend
  • Intensity slider
  • 5 emotions
Open demo
Coming soon

Emo-FiLM Keyframes

Word-level emotion control with FiLM modulation

  • Per-word emotion
  • Timeline editor
  • Smooth transitions
Coming soon

DrawSpeech

Sketch pitch and energy curves directly

  • Draw pitch contour
  • Draw energy
  • Canvas editor
Coming soon

Chatterbox

Single-parameter emotion exaggeration (0=monotone, 2=dramatic)

  • One slider
  • Paralinguistic tags
  • Quick testing
Coming soon

MaskGCT Prosody

Masked generative codec for prosody editing

  • Prosody masking
  • Selective regeneration
  • Fine control

Demo pages are auto-generated by research agents using Codex.

See .claude/commands/create-demo-page.md for the template.