You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Add optional local live subtitles in the TUI, with SenseVoice Cantonese and
Qwen3 model presets, explicit verified downloads, and scrollable session
history. Speech uses an isolated runtime and a separate stream connection;
subtitles may lag behind or differ in timing from playback. Model downloads
support a configurable mirror, disk space checks, and official fallback
while retaining fixed SHA-256 verification.
Bound subtitle reconnects until speech production stays stable for a minute,
preserve Chinese captions in short terminals, and make persistent speech
failures available through ? and the activity log. Keep subtitle popups
and panel layouts consistent across terminal resizes; suppress track artwork
probes throughout enabled subtitle sessions. Support Unicode transcripts
under legacy locales, improve model diagnostics, and avoid repeated work
while subtitles are disabled or waiting for playback.
Reuse model verification for unchanged files within a TUI session to avoid
hashing large models on every subtitle reconnect. Explicit verification,
diagnostics, and installation retain full SHA-256 checks.
Keep receiving track updates when mpv sends invalid UTF-8 in metadata,
preserving readable Unicode and recoverable legacy title bytes.
Show a replacement character for truncated UTF-8 in Chinese ICY titles
instead of displaying the incomplete bytes as Latin-1 characters.
Recover GBK ICY titles containing valid UTF-8 subsequences through mpv IPC,
including names such as 张国荣 and 河图.
Add an opt-in asiafm_metadata_patch setting to remove the fixed Cantonese
station watermark and clarify damaged track fields without guessing songs.