HandshakePrice

(SN58)

Dippy Speech (SN58) is a specialized subnet on the Bittensor network dedicated to building advanced, open-source empathetic speech models for AI companionship. While other subnets focus on text generation, SN58 specifically develops Text-to-Speech (TTS) technology. Its goal is to create lifelike, human-sounding voices that power the user experience in the Dippy application, making AI interactions feel more natural and emotionally resonant.

As part of the transition to Dynamic TAO (dTAO), the SN58 token allows users and investors to participate directly in the subnet's ecosystem. Through the dTAO mechanism, individuals can stake on specific subnets like SN58, allowing for decentralized voting and governance. This system ensures that the success and quality of the speech models developed by the subnet translate into value within the network's specialized tokenomics.

Dippy distinguishes itself by taking a much less censored stance compared to mainstream AI assistants. The project focuses on providing an open environment for empathetic roleplay and companionship, avoiding the heavy filters often found on centralized platforms. This commitment to creative freedom is a primary draw for its community, allowing for more authentic and diverse interactions with AI characters.

Miners on SN58 utilize Parler-TTS Mini v1 as their base reference model. To participate, they must provide GPU compute power to handle text-to-speech benchmarking and training. The development pipeline involves fine-tuning these models and submitting weights to a shared pool in Safetensors format. This competitive environment incentivizes miners to produce the highest quality audio output to earn rewards.

The subnet employs rigorous scoring metrics to ensure the models meet high standards. Evaluation is based on the Word Error Rate (WER), which tracks accuracy, and a Human Likeness Score, which measures how natural and empathetic the voice sounds. These metrics allow the decentralized network to objectively rank the performance of different models and reward the developers who produce the most lifelike speech technology.

The project is currently in a phased rollout. According to the roadmap, the most anticipated milestone is Phase 3, which marks the full end-to-end integration of SN58’s decentralized voice models into the consumer app. Once complete, users will interact with AI characters powered by the lifelike speech technology developed on the Bittensor network. Additionally, as these models are open-source, they may eventually be used in other third-party applications and games.