Skip to content

VOICE_COMMANDS.md

Latest commit

 

History

History
81 lines (53 loc) · 4.19 KB

VOICE_COMMANDS.md

File metadata and controls

81 lines (53 loc) · 4.19 KB

SwiftFloris Voice Commands

SwiftFloris stores voice command definitions locally and can parse command-like transcripts with fuzzy matching. Dictation is currently handed to FUTO Voice Input as an external Android voice keyboard, so live command execution depends on a future transcript handoff path from that external IME.

Built-In Commands

Spoken phrase Aliases Action
delete that delete it, remove that Deletes the previous word, or the active selection when the editor provides one.
undo Requests undo from the active editor.
redo Requests redo from the active editor.
select all select everything Selects all editable text in the active field.
clear text clear field, delete all Selects the active field and clears its text.
new paragraph next paragraph Inserts a blank line between paragraphs.
new line line break Inserts a line break.
capitalize next word capitalize next Enables one-shot capitalization for the next typed word.
go to start go to beginning, start of field Moves the cursor to the beginning of the field.
go to end end of field Moves the cursor to the end of the field.

Recognition Behavior

SwiftFloris normalizes command phrases before matching:

  • Case and punctuation are ignored.
  • Diacritics are ignored, so accented recognition output can still match English commands.
  • Leading or trailing please is ignored.
  • Small recognition errors are tolerated through fuzzy matching.

Confidence handling:

  • 0.85 and above: execute the matched command.
  • 0.50 to 0.85: hold a suggestion so the UI can offer accept or reject.
  • Below 0.50: treat the transcript as normal dictated text and insert it.

Unrecognized command-like speech is intentionally inserted as text. This avoids destructive false positives when a user dictates ordinary sentences such as delete the old note after lunch.

Custom Commands

Custom commands are managed in Settings > Voice input > Voice commands.

Each custom command has:

  • A spoken phrase.
  • A mapped command action.
  • An enabled or disabled state.

Custom phrases are stored on device in SwiftFloris preferences. Disabled commands are ignored by the parser and can be re-enabled later.

Errors And Recovery

FUTO Voice Input is offline-first, but SwiftFloris still models timeout and failure paths so future transcript integrations can fail safely.

Condition SwiftFloris behavior
No transcript or no match Keeps voice input ready and inserts nothing.
Unrecognized transcript Inserts the transcript as normal text.
Low-confidence command Returns a pending suggestion for accept or reject.
Timeout or recognizer busy Marks the attempt retryable without deleting or inserting text.
FUTO missing, disabled, or missing microphone permission Shows setup-oriented unavailable state.

Accent FAQ

What if my accent is not recognized?

Start by checking the phrase shown by FUTO. If FUTO hears a close variant, SwiftFloris fuzzy matching may still detect the command. If FUTO consistently hears a different phrase, add that phrase as a custom command in Voice input settings.

Can I use non-English command phrases?

Yes, as custom commands. Built-in command phrases are currently English, but custom commands can map any phrase FUTO can transcribe to a supported action.

Why did my command get typed instead of executed?

SwiftFloris only executes high-confidence command matches. Low-confidence matches should be confirmed by the user, and weak matches are inserted as text to avoid accidental destructive actions.

Why do commands not run immediately after FUTO dictation?

SwiftFloris currently launches FUTO as a separate Android voice keyboard. Android does not provide SwiftFloris with FUTO's dictated transcript during that handoff, so parser and executor support is ready but live command execution still needs a transcript integration path.

Validation Status

The parser, custom-command storage, command executor, fallback handler, and settings UI are covered by unit tests. Human validation across diverse accents is still required before claiming production recognition accuracy.