Загрузка видео...
Не удалось загрузить видео
The new open-source Text to Speech model: Fish Speech 1.4 is brilliant! Trained on a massive 700K hours of multilingual speech data in 8 languages - Instant voice cloning 🗣️ - Ultra-low latency ⚡ - Compact model (~1GB weights) 🏋️♂️
228,836 просмотров • 1 год назад •via X (Twitter)
Комментарии: 8

GitHub: HuggingFace Model:

Will somebody please stuff this into a browser so my phone can read things to me?

Cool idea

Supported Languages

Hmmmm license is so so... Only advantage over CosyVoice is that there more language supported for the moment. Still will be using Cosyvoice for moment.

👍

Can we quantize this to fit on an iPhone 🔥

Yes, we should be able to apply Vector Quantization or DQR (Dynamic Quantized Representations). Another alternative would be MELLE for preserving fidelity. MELLE uses continuous-valued tokens and a regression loss function to model the probability distribution of these tokens, achieving superior performance and robustness compared to models using discrete codes

