Initial roadmap

[unilight]

Datasets

• JSUT
• JVS
• Hi-Fi-Captain
• CSJ
• Emilia-YODAS jp

Models

• FastSpeech 2 (w/ julius forced alignment) + Hi-Fi GAN / PWG
• VITS w/ mel output (i.e., w/o GAN) + Hi-Fi GAN / PWG
• Matcha-TTS w/ julius forced alignment + Hi-Fi GAN / PWG
• Matcha-TTS w/ MAS + Hi-Fi GAN / PWG
• VALL-E
• E2-TTS

Refactor

• Apply accelerate to every recipe?
  • Applied recipes:
    • egs/hificaptain_jp_female/tts2
    • egs/hificaptain_jp_female/tts3

	JSUT	JVS	Hi-Fi Captain	CSJ	Emilia-YODAS jp
FastSpeech2	2025/1/30	2025/1/24	2025/2/3	N/A	N/A
VITS	2025/4/15		2025/4/15	N/A	N/A
MatchaTTS w/ julius forced alignment	2025/2/12	2025/2/12	2025/2/12	N/A	N/A
MatchaTTS w/ MAS	2025/4/10	2025/4/15	2025/4/10	N/A	N/A
VALL-E			2025/4/18
E2-TTS			2025/4/15 (basically fails)