Audio & Recordings 音訊與錄音

Settings live in Settings → Audio → Microphone. The dropdown lists every input device the OS reports plus a System Default option (this is what TypeLate picks when nothing else is set).

• Choosing a specific device by name pins recordings to it.
• The Refresh button re-enumerates devices — useful after plugging in a USB mic or pairing Bluetooth headphones.
• If the chosen device is unplugged before you record, TypeLate falls back to the system default silently.
• If it’s unplugged mid-recording, the recording stops with an error.

In Settings → Application → Mute on Recording (default on).

When enabled, TypeLate mutes the system output while you record and restores it afterwards. This stops your speakers’ output (music, video calls, notifications) from leaking into the mic and making transcription worse.

Implementation notes worth knowing:

• This is system-wide muting — every app’s output is silenced during recording, not just the active one.
• If your system was already muted before recording, it stays muted afterwards (TypeLate doesn’t unmute what it didn’t mute).
• If muting fails (rare), recording continues without warning — you might just hear yourself echoed.

In Settings → Audio → Sound Theme you can pick one of five presets. Each maps four “slots” — start of recording, stop of recording, error, and word-learned — to a sound:

Theme	Start	Stop	Error	Learned
Default	Funk	Bottle	Ping	Glass
Gentle	Tink	Pop	Basso	Purr
Minimal	Pop	Pop	Basso	Pop
Retro	Morse	Submarine	Sosumi	Hero
Custom	(your file)	(your file)	(your file)	(your file)

On macOS the built-in sounds map to /System/Library/Sounds/{name}.aiff. On Windows, start/stop are bundled WAV files played via PlaySound.

Custom sounds

Pick Custom and you can wire up any combination of .wav, .mp3, .aiff, or .m4a files. Each slot is independent — you can replace just Start and leave the rest default.

The Preview button next to each slot plays the currently-assigned sound at TypeLate’s effective volume.

The audio file behind every transcription is normally kept on disk — handy if you want to play it back or re-transcribe it. The retention policy in Settings → Recording Storage → Audio Retention decides how long.

Option	Behaviour
Keep Forever	Never auto-delete.
30 Days (default)	Files older than 30 days are removed on next launch.
14 Days	Same as above, 14 days.
7 Days	Same as above, 7 days.
Don’t Keep	Audio file deleted immediately after transcription.

Two things to know:

• Cleanup uses the file’s modification time, not the transcription record’s creation time. Don’t manually touch files inside the recordings folder.
• Switching to Don’t Keep does not delete the SQLite history record — you’ll still see the entry in History, just without playback or re-transcribe.

Platform	Path
macOS	~/Library/Application Support/com.bobo.typelate/AppData/recordings/
Windows	%AppData%\com.bobo.typelate\AppData\recordings\

Use Open Folder in Settings → Recording Storage to jump straight there. The same panel shows total disk usage and file count.

The History view in the dashboard shows all transcriptions, newest first, grouped as Today, Yesterday, or MMM D by date.

Each row shows:

• A truncated transcript preview (first 60 characters).
• A small AI-enhanced indicator dot if the LLM step was applied.
• A Failed badge if the transcription errored out.
• The time (HH:MM), recording duration (2s, 1m 34s…), and character count.

The list paginates with infinite scroll — 20 items per page.

Search

The search box at the top filters by raw transcription and the enhanced result. It’s a literal substring match (case-sensitive, no special operators). SQL wildcards in the search input are escaped, not interpreted.

Export

The Export button copies a JSON array of every transcription to your clipboard, including: id, timestamp, ISO date, raw text, processed text, character count, recording duration, model IDs, and the wasEnhanced flag.

Click the play button on any history row to hear the original audio inline. The button morphs into a stop icon during playback; click again to pause. Audio plays through the standard system output.

There are no keyboard shortcuts and no scrubbing UI — playback is meant for spot-checking, not editing.

If the file’s gone (retention deleted it, or you cleared the recordings folder), you’ll see Recording file not available under the row for 3 seconds.

Open a history row and click Re-transcribe (only available if the audio file still exists and the original transcription succeeded). TypeLate runs the audio again using your current global settings:

• Selected Whisper model
• Selected LLM model
• Prompt mode and custom prompt
• Enhancement threshold setting
• Transcription language
• Top 50 dictionary terms by weight

You can’t override these per-record — re-transcribing is “what would TypeLate produce today, with the same audio?”.

The button is a two-click confirm (3-second window) to prevent accidents. After it succeeds, the row’s text and metadata are overwritten — there’s no version history.

In Settings → Recording Storage → Delete All Recordings there’s a button that wipes every audio file in the recordings folder after a confirm dialog. History records are preserved — you’ll still see what was said, you just can’t play or re-transcribe any of them.

To delete a specific row’s audio + history together, use the per-row delete in History (also two-click confirm). That removes the SQLite record, the audio file, and the related entries in the cost-tracking table.