Ant Group's Ming-3.0-tts fits top-tier Text to Speech quality into just 0.5B parameters, and it is MIT licensed and punches well above its weights
It is the only model under 1B parameters in the top 3 of any of our Text to Speech leaderboards. Every other open weights TTS model in the top 10 of our Controlled Voice Arena is 3.5B or larger, so @AntLingAGI's Ming-3.0-tts reaches the same tier with 7x to 70x fewer parameters.
Links below 🧵