Mehta et al. · 2024 · CFM
A compact acoustic model that uses conditional flow matching instead of a long diffusion chain. Fast, small, and a clean explanation of why flow matching showed up in F5 and CosyVoice.
The short paper you read before you train a flow TTS.
Vendor / paper demo
LJSpeech test (MAT-10, sentence 1) · Matcha-TTS official demo, stimuli/sample_from_test/MAT-10_1.wav.
Matcha-TTS paper demo, MAT-10 sentence 1 from the test set. The demo page loads transcripts in the player; this file is that stimulus.