Running a Miner

Hardware Recommendations

• GPU: RTX 3090+ for multiple models
• RAM: 32GB+
• Storage: Fast SSD for model caching

Model Choices

• XTTS → Fast inference, multilingual.
• RVC → High fidelity cloning.
• Bark → Expressive, creative voices.
• Tortoise → Slow but ultra-realistic.

Example Setup with XTTS

1. Install dependencies:

bash

pip install TTS torch torchaudio

2. Basic miner template (extend subnet template when released):

python

import bittensor as bt
from TTS.api import TTS
 
class EchoIcMiner(bt.dendrite):
    def __init__(self):
        super().__init__()
        self.model = TTS("tts_models/multilingual/multi-dataset/xtts_v2").cuda()
 
    def forward(self, task):
        # Download reference audio
        ref_wav = download(task['voice_url'])
        output_wav = self.model.tts_to_file(
            text=task['transcript_text'],
            speaker_wav=ref_wav,
            language="en"  # Detect/auto
        )
        uploaded_url = upload_to_ipfs(output_wav)
        return {"output_url": uploaded_url}
 
miner = EchoIcMiner()
miner.run()

Tips for High Scores

• Pre-load models to beat timeouts.
• Clean reference audio (6-30s, no noise).
• Balance speed vs quality.

← Previous Getting Started Next → Running a Validator