README.md

metadata

tags:
  - rvc
  - voice-conversion
  - quran
  - arabic
  - vintage
  - ai-voice
license: mit
datasets:
  - elgohary47/sheikh_refaat_dataset
language:
  - ar
base_model:
  - IAHispano/Applio
pipeline_tag: audio-to-audio

Sheikh Muhammad Refaat (RVC v2) - 250 Epochs

This is a high-quality RVC v2 model of Sheikh Muhammad Refaat, trained on 30 minutes of restored vintage audio.

Usage Recommendations

To get the authentic "Vintage Radio" sound, use these settings in RVC/Applio:

• Pitch (Transpose):
  • -2 or -3 (If your voice is higher/younger).
  • 0 (If using a professional imitator as input).
• Index Rate: 0.75 (Critical for the grainy texture).
• Resample: 0 (Do not change).
• Protect Voiceless Consonants: 0.33 (Keeps breathing natural).
• Hop Length: 128 (Better for slow recitation).

Pipeline for Best Results

Since Sheikh Refaat has a unique Maqam style, do not use your own voice directly.

1. Find a recording of Sheikh Muhammad Omran (imitating Refaat).
2. Remove reverb using UVR5.
3. Use that clean audio as the input for this model.

Files Included

• .pth: The model weights (250 Epochs).
• .index: The feature index (Essential for texture).

Trained by: elgohary47 Dataset: Restored 1940s Radio Archives(elgohary47/sheikh_refaat_dataset)