1. Introduction

MyShell TTS Subnet: Decentralizing Open-Source TTS Innovation

MyShell TTS Subnet is a decentralized ecosystem driving cutting-edge open-source text-to-speech (TTS) model development. Powered by the Bittensor blockchain and a unique incentive system, we're pushing the boundaries of TTS possibilities.

In this subnet, MyShell's recently open-sourced TTS model, MeloTTS, serves as our backbone - a state-of-the-art multi-lingual model leveraging transformers to encode input into phoneme latent space, then using normalizing flow and transposed convolution to decode into realistic speech. Despite running on CPUs, its efficient architecture enables real-time inference.

2. How Does It Work?

2.1 Evaluation Protocol

Our protocol assesses TTS performance across key metrics:

Naturalness: How human-like the generated speech sounds

Intelligibility: Clarity and understandability

Expressiveness: Ability to convey emotions and prosody

Consistency: Stability across inputs and speakers

We employ a custom rater model to extract voice features and calculate these metrics, ensuring the best models are identified and rewarded - continuously elevating open-source TTS quality.

2.2 Decentralized Incentives

Miners: Skilled developers training and submitting TTS models to a shared Hugging Face pool, with metadata on the Bittensor chain.

Validators: Experts generate test cases, assess model performance via our protocol, and assign weights accordingly.

Bittensor Blockchain: Decentralized ledger aggregating validator weights using Yuma Consensus to determine final miner weights and emissions, rewarding top contributions.

This incentive design motivates miners to create exceptional models while encouraging validators to provide accurate assessments - fostering a self-sustaining ecosystem pioneering open-source TTS innovation.

3. The vision of MyShell TTS Subnet

Incentivizing open-source contributions by rewarding creators and users for building AI modules and optimizing datasets through Bittensor's incentive mechanism.

We're bridging the gap between the decentralized world, AI, and consumers while unleashing the power of open-source communities to contribute AI models. By collaborating with Bittensor to build the MyShell TTS Subnet, our vision is threefold:

1. Provide real-world use cases for the largest decentralized AI training network by onboarding MyShell's million users on our platform on Bittensor.

1. Continuously developing and refining the MyShell TTS Subnet itself, fostering a self-sustaining cycle of innovation driven by the incentivized community.