---
license: apache-2.0
tags:
  - voice-changer
  - rvc
  - spin
  - f0-categorized
  - non-speech
  - pretrained-model
datasets:
  - custom
language:
  - multilingual
---

# 🔊 KLM BlackTone Large (Pretrained Voice Changer Model)

**KLM BlackTone Large** is a pretrained model for real-time **voice conversion**, built using the **Spin Embedder**. It was trained on a massive dataset of over **11,000 hours** from **595 speakers**, carefully categorized by **F0 range**

---

## 🚀 Requirements

To use this model **without errors**, you **must** use **Applio v3.2.9 or later**, specifically versions **updated after July 15**.  
Even if your Applio version is labeled 3.2.9, models updated **before** July 15 may still result in `mismatch` errors.

---

## 🌟 Key Features

- ✅ **595 F0-Categorized Speakers**
- ✅ **11,000+ hours of curated training data**
- ✅ **Spin Embedder** for flexible and generalizable voice transfer
- ✅ Real-time inference support
- ✅ Extensive support for **non-verbal sounds** such as:
  - **Coughs**
  - **Laughter**
  - **Whispers**
  - And other expressive human vocal behaviors

These features are made possible by including a large proportion of **non-speech data** in the training set.

---

## 📊 Recommended Usage

To fully utilize BlackTone’s capabilities—especially non-speech inference—you should include **10–20% non-speech data** in your fine-tuning dataset.

If you **do not** have such data, you can try setting the **Feature Index** to `0`, which may enable limited inference of non-verbal sounds. However, **non-speech training data is highly recommended** for best results.

---

## 📎 License

Apache 2.0

---

## 📫 Contact

For usage reports, contributions, or collaborations, please open an issue or contact the model maintainer.