Orpheus-3B-ASMR / README.md
nyuuzyou's picture
Super-squash branch 'main' using huggingface_hub
b6c3f2a verified
metadata
language:
  - en
pipeline_tag: text-to-speech
license: apache-2.0
base_model: unsloth/orpheus-3b-0.1-ft
datasets:
  - nyuuzyou/asmr
tags:
  - asmr
  - lora
co2_eq_emissions:
  emissions: 1280
  source: Calculated based on power consumption and regional carbon intensity
  training_type: fine-tuning
  geographical_location: Chelyabinsk, Russia
  hardware_used: 1 RTX 4090 GPU

Orpheus 3B ASMR (Merged)

Orpheus 3B model fine-tuned on ASMR data and merged for improved soft-spoken speech generation.

Model Details

  • Base: unsloth/orpheus-3b-0.1-ft
  • Training Data: nyuuzyou/asmr (283K clips, 307 hours)
  • Training: 170,000 steps (~40 hours on RTX 4090)
  • Type: Merged model (LoRA + base weights)

Capabilities

  • Enhanced soft-spoken speech on pre-trained voices (e.g., "tara")
  • Improved gentle vocal characteristics
  • Standard Orpheus features: voice cloning, emotion control, streaming

Limitations

  • Cannot generate true whispering - training method insufficient for complex whisper synthesis
  • Limited ASMR authenticity - does not produce human-like ASMR content
  • Best results with existing voice profiles rather than novel ASMR voices

Ethics

Do not use for impersonation without consent or deceptive purposes.