Fastest and only macOS Dictation app with on-device STT and custom trained AI enhancement model. Windows, iOS testflight! Android, Linux waitlist open! A local Wispr Flow alternative. Follow https://x.com/fluidvoiceapp for more
import Foundation
/// Bounded, presentation-only word/punctuation comparison. Whitespace is preserved verbatim.
struct HistoryTextDiff: Sendable {
struct Run: Equatable, Sendable {
var text: String
let changed: Bool
}
let original: [Run]
let final: [Run]
var hasChanges: Bool { self.original.contains { $0.changed } || self.final.contains { $0.changed } }
static func compare(original: String, final: String) -> HistoryTextDiff? {
guard original.utf8.count + final.utf8.count <= 24_000 else { return nil }
let pattern = #"\s+|[\p{L}\p{N}_]+|[^\s\p{L}\p{N}_]"#
guard let expression = try? NSRegularExpression(pattern: pattern) else { return nil }
func tokens(_ text: String) -> [String] {
expression.matches(in: text, range: NSRange(text.startIndex..., in: text)).compactMap {
Range($0.range, in: text).map { String(text[$0]) }
}
}
let before = tokens(original)
let after = tokens(final)
guard before.count + after.count <= 2048 else { return nil }
let changes = after.difference(from: before)
var removed = Set<Int>()
var added = Set<Int>()
for change in changes {
switch change {
case let .remove(offset, _, _): removed.insert(offset)
case let .insert(offset, _, _): added.insert(offset)
}
}
func runs(_ tokens: [String], changed: Set<Int>) -> [Run] {
var result: [Run] = []
for (index, token) in tokens.enumerated() {
let isChanged = changed.contains(index)
if result.last?.changed == isChanged {
result[result.count - 1].text += token
} else {
result.append(Run(text: token, changed: isChanged))
}
}
return result
}
return HistoryTextDiff(original: runs(before, changed: removed), final: runs(after, changed: added))
}
}