Echo Live includes 43 DSP voice presets. But for hyper-realistic voice conversion, you can import RVC neural voice models trained by the open-source community.
Find a voice model from Hugging Face, Weights.gg, or the AI Hub Discord. Download the .pth file to your computer.
Launch the app and switch to the "RVC Voices" tab in the voice selector. This is separate from the DSP voices tab.
Drag and drop the .pth file into the import zone, or click "Import Model" and select the file. The model loads in seconds.
Tune the pitch offset if needed (e.g., +4 for male-to-female). The app auto-detects optimal settings, but you can fine-tune via the calibration panel.
Select your microphone, enable voice conversion, and speak. Your voice is converted in real-time with GPU-accelerated inference.
The largest open-source model repository. Search for "RVC" to find thousands of community-trained voice models.
A dedicated RVC model sharing platform with previews, ratings, and easy downloads.
The largest RVC community Discord server. Model sharing, training help, and technical discussion.
Reddit community for RVC discussion, model sharing, and troubleshooting.
Want a voice that does not exist in the community? You can train your own RVC model using free tools. The process requires a GPU with at least 4GB VRAM and takes 1-4 hours depending on your hardware.
Start by collecting 10-30 minutes of clean, isolated vocal audio. The source audio should be free of background music, noise, and reverb. Audiobook recordings, podcast episodes, and clean interview audio work well.
The most popular training tools are RVC WebUI (runs locally on your GPU) and Google Colab notebooks (runs in the cloud for free). Both produce .pth model files that are directly compatible with Echo Live.
For best results, use the RVC v2 architecture with a 48kHz sample rate and train for 200-400 epochs. Monitor the training loss — if it plateaus, you can stop early. Over-training can actually reduce quality.
Voice cloning technology is powerful and comes with responsibility. Always get explicit consent before creating a model of someone else's voice. Never use voice cloning for fraud, impersonation, or deception. Creating models of fictional characters, your own voice, or consenting participants is generally acceptable. When in doubt, ask.