4c086251d9
Native editor to fix speaker-ID errors after transcription (modeled on recap-relay's correction UX): rename a speaker in the legend, merge two speakers, or reassign an individual transcript line. Saving rewrites speakers.json, re-renders transcript.md + recap.html, and updates the voiceprint memory — so a correction compounds: naming an "Unknown" speaker teaches that voice for future calls. - SpeakerEditing (pure, tested): replaceSpeaker (rename = merge-onto-existing), reassign, netNameMap (compose ops), and remap (apply a name map to a recap's structured fields + whole-word free text, so summaries/extras update without re-LLM). - RecapEditModel (@MainActor): loads speakers.json (+ optional recap.json + cluster_fingerprints.json); on save writes the resolved speakers.json, re-renders, and reconciles voiceprints — merge keeps the survivor's print; rename/name-an-Unknown enrolls the cluster's fingerprint under the new name. - TranscriptEditorView (SwiftUI) + EditorWindow (AppKit window for the LSUIElement app); menu gains "Edit speakers". - Pipeline now persists cluster_fingerprints.json (every cluster incl. Unknown) and recap.json (RecapFile) so the editor can learn voices + re-render offline. - RecapModels made Codable; TranscriptAssembler exposes allFingerprints; VoiceprintStore gains enroll() + merge(). 52/52 XCTest (6 new, incl. a full rename→artifacts→voiceprint round-trip on disk).
129 lines
4.7 KiB
Swift
129 lines
4.7 KiB
Swift
import Foundation
|
|
|
|
/// Local persistence of named voiceprints — the compounding-identity layer.
|
|
///
|
|
/// File `~/Ten31Transcripts/voiceprints.json`:
|
|
/// `{ "<name>": { "vector": [192 floats], "updated": <iso>, "calls": <int> } }`
|
|
///
|
|
/// On send → `knownVoiceprints()` feeds `label-merge`. On response → `update(with:)`
|
|
/// stores/refreshes vectors for speakers resolved by **visual** (overlap ≥ ~0.8)
|
|
/// or **voiceprint** match. Never stores `Unknown_N` / `Speaker_unknown`.
|
|
///
|
|
/// Thread-safe (lock-guarded); the sequential pipeline is the only writer.
|
|
final class VoiceprintStore {
|
|
struct Entry: Codable, Equatable {
|
|
var vector: [Float]
|
|
var updated: String
|
|
var calls: Int
|
|
}
|
|
|
|
private let url: URL
|
|
private let minOverlapToStore: Double
|
|
private let lock = NSLock()
|
|
private var entriesStore: [String: Entry] = [:]
|
|
|
|
init(fileURL: URL, minOverlapToStore: Double = 0.8) {
|
|
self.url = fileURL
|
|
self.minOverlapToStore = minOverlapToStore
|
|
load()
|
|
}
|
|
|
|
var entries: [String: Entry] {
|
|
lock.lock(); defer { lock.unlock() }
|
|
return entriesStore
|
|
}
|
|
|
|
/// Vectors keyed by name, for the `known_voiceprints` field.
|
|
func knownVoiceprints() -> [String: [Float]] {
|
|
lock.lock(); defer { lock.unlock() }
|
|
return entriesStore.mapValues { $0.vector }
|
|
}
|
|
|
|
/// Persist fingerprints from a `label-merge` response for confidently-named
|
|
/// speakers only.
|
|
func update(with response: LabelMergeResponse) {
|
|
lock.lock(); defer { lock.unlock() }
|
|
let now = ISO8601DateFormatter().string(from: Date())
|
|
for sp in response.speakers {
|
|
guard !Self.isUnknown(sp.name) else { continue }
|
|
let acceptable: Bool
|
|
switch sp.source {
|
|
case "mic_channel": acceptable = true // the user's own clean mic voiceprint
|
|
case "visual": acceptable = (sp.overlapConfidence ?? 0) >= minOverlapToStore
|
|
case "voiceprint": acceptable = true // already matched a known print
|
|
default: acceptable = false // unmatched
|
|
}
|
|
guard acceptable, let vector = sp.fingerprint ?? response.fingerprints[sp.name],
|
|
!vector.isEmpty else { continue }
|
|
var entry = entriesStore[sp.name] ?? Entry(vector: vector, updated: now, calls: 0)
|
|
entry.vector = vector
|
|
entry.updated = now
|
|
entry.calls += 1
|
|
entriesStore[sp.name] = entry
|
|
}
|
|
save()
|
|
}
|
|
|
|
func rename(_ old: String, to new: String) {
|
|
lock.lock(); defer { lock.unlock() }
|
|
guard old != new, let e = entriesStore.removeValue(forKey: old) else { return }
|
|
entriesStore[new] = e
|
|
save()
|
|
}
|
|
|
|
/// Enroll/refresh a voiceprint under `name` (e.g. after the user renames an
|
|
/// "Unknown" speaker to a real name — we learn that voice for future calls).
|
|
func enroll(name: String, vector: [Float]) {
|
|
guard !name.isEmpty, !Self.isUnknown(name), !vector.isEmpty else { return }
|
|
lock.lock(); defer { lock.unlock() }
|
|
let now = ISO8601DateFormatter().string(from: Date())
|
|
var entry = entriesStore[name] ?? Entry(vector: vector, updated: now, calls: 0)
|
|
entry.vector = vector
|
|
entry.updated = now
|
|
entry.calls += 1
|
|
entriesStore[name] = entry
|
|
save()
|
|
}
|
|
|
|
/// Merge `absorbed` into `survivor`: drop the absorbed entry, keep the survivor's
|
|
/// print (the user said they're the same person).
|
|
func merge(_ absorbed: String, into survivor: String) {
|
|
lock.lock(); defer { lock.unlock() }
|
|
guard absorbed != survivor else { return }
|
|
entriesStore.removeValue(forKey: absorbed)
|
|
save()
|
|
}
|
|
|
|
func remove(_ name: String) {
|
|
lock.lock(); defer { lock.unlock() }
|
|
entriesStore.removeValue(forKey: name)
|
|
save()
|
|
}
|
|
|
|
func reset() {
|
|
lock.lock(); defer { lock.unlock() }
|
|
entriesStore = [:]
|
|
save()
|
|
}
|
|
|
|
// MARK: - Persistence (call with lock held)
|
|
|
|
private func load() {
|
|
guard let data = try? Data(contentsOf: url),
|
|
let decoded = try? JSONDecoder().decode([String: Entry].self, from: data) else { return }
|
|
entriesStore = decoded
|
|
}
|
|
|
|
private func save() {
|
|
let encoder = JSONEncoder()
|
|
encoder.outputFormatting = [.prettyPrinted, .sortedKeys]
|
|
try? FileManager.default.createDirectory(at: url.deletingLastPathComponent(),
|
|
withIntermediateDirectories: true)
|
|
if let data = try? encoder.encode(entriesStore) { try? data.write(to: url) }
|
|
}
|
|
|
|
private static func isUnknown(_ name: String) -> Bool {
|
|
LabelMergeResponse.isUnknownName(name)
|
|
}
|
|
}
|