ten31-transcripts/Ten31Transcripts/Session/TranscriptAssembler.swift

import Foundation

/// Concatenates per-chunk `label-merge` results into one global `speakers.json`:
/// segment times offset back to global seconds, speakers unified across chunks by
/// name, and fingerprints collected for the voiceprint store.
enum TranscriptAssembler {
    struct ChunkResult {
        let chunkStart: Double           // global seconds
        let response: LabelMergeResponse
    }

    struct Assembled {
        let speakersFile: SpeakersFile
        let fingerprints: [String: [Float]]   // name -> 192-dim, for VoiceprintStore
    }

    /// Source ranking when the same name appears across chunks with different sources.
    private static func rank(_ source: String) -> Int {
        switch source {
        case "visual": return 3
        case "voiceprint": return 2
        default: return 1            // unmatched
        }
    }

    private static func isUnknown(_ name: String) -> Bool {
        LabelMergeResponse.isUnknownName(name)
    }

    static func assemble(sessionId: String, app: String, chunks: [ChunkResult]) -> Assembled {
        var segments: [SpeakersFile.Segment] = []
        var bestSpeaker: [String: SpeakersFile.Speaker] = [:]
        var fingerprints: [String: [Float]] = [:]
        var models: [String: String] = [:]
        var duration = 0.0

        for chunk in chunks {
            let offset = chunk.chunkStart
            // Audio length from the chunk window, so silent/all-unknown calls still
            // report a real duration (not just the last segment's end).
            duration = max(duration, offset + chunk.response.duration)

            for seg in chunk.response.segments {
                let start = seg.startSeconds + offset
                let end = seg.endSeconds + offset
                segments.append(.init(start: start, end: end, speaker: seg.speaker, text: seg.text))
                duration = max(duration, end)
            }

            for sp in chunk.response.speakers {
                let candidate = SpeakersFile.Speaker(
                    name: sp.name, source: sp.source,
                    overlapConfidence: sp.overlapConfidence, matchSimilarity: sp.matchSimilarity)
                if let existing = bestSpeaker[sp.name] {
                    if rank(sp.source) > rank(existing.source) { bestSpeaker[sp.name] = candidate }
                } else {
                    bestSpeaker[sp.name] = candidate
                }
                // Collect named fingerprints only (never Unknown_N / Speaker_unknown).
                if !isUnknown(sp.name), let fp = sp.fingerprint, fp.count > 0 {
                    fingerprints[sp.name] = fp
                }
            }
            for (name, fp) in chunk.response.fingerprints where !isUnknown(name) && fp.count > 0 {
                fingerprints[name] = fp
            }
        }

        segments.sort { $0.start < $1.start }
        let speakers = bestSpeaker.values.sorted { $0.name < $1.name }
        models = chunks.last?.response.models ?? [:]

        let file = SpeakersFile(
            sessionId: sessionId, app: app, durationSec: duration,
            speakers: speakers, segments: segments, models: models)
        return Assembled(speakersFile: file, fingerprints: fingerprints)
    }
}