Chunk overlap + overlap-aware stitching

Chunks were contiguous (start = prev end) with a naïve offset-concat stitch — no
overlap. That cut sentences at boundaries, denied the diarizer context at edges, and
let one voice split across chunks (the MH/Unknown_0 problem). Now each ~150s body is
sliced with a 15s margin on both sides ([bodyStart-15, bodyEnd+15]); the stitcher
keeps a segment only in the chunk that owns its MIDPOINT (body region) and drops it
from the neighbour's margin — so boundary-spanning speech is seen whole by the
backend and kept exactly once.

- SessionPackager.PlannedChunk gains bodyStart/bodyEnd; planChunks adds overlapSeconds.
- TranscriptAssembler.ChunkResult carries body bounds (defaults keep-all → no-overlap
  behaviour preserved for existing callers); assemble dedups by midpoint-in-body.
- TranscriptPipeline passes body bounds through.

Complements (doesn't replace) the fragment-smoothing + reconciliation safety nets;
this is the upstream fix. ~+20% backend audio per interior chunk. 63/63 XCTest
(new: overlap window layout + boundary-segment dedup).
This commit is contained in:
Grant Gilliam
2026-06-08 13:03:56 -05:00
parent 1c133c8970
commit ab910cf742
4 changed files with 60 additions and 18 deletions
+20 -9
View File
@@ -10,24 +10,35 @@ import AVFoundation
enum SessionPackager {
struct PlannedChunk: Equatable {
let index: Int
let start: Double // global seconds
let end: Double
let start: Double // sliced window start (global seconds, incl. overlap margin)
let end: Double // sliced window end (incl. overlap margin)
let bodyStart: Double // the region this chunk OWNS (no overlap) for stitch dedup
let bodyEnd: Double
}
/// One chunk if short; otherwise even ~`chunkSeconds` windows.
/// One chunk if short; otherwise ~`chunkSeconds` bodies, each sliced with an
/// `overlapSeconds` margin on both sides. The margin gives the backend context at
/// boundaries (so a sentence isn't cut and the diarizer attributes edge speech
/// correctly and keeps a voice consistent across chunks); the stitcher keeps only
/// each chunk's owned `body` region, deduping the overlap.
static func planChunks(durationSec: Double,
chunkSeconds: Double = 150,
overlapSeconds: Double = 15,
thresholdSec: Double = 180) -> [PlannedChunk] {
guard durationSec > thresholdSec else {
return [PlannedChunk(index: 0, start: 0, end: durationSec)]
return [PlannedChunk(index: 0, start: 0, end: durationSec, bodyStart: 0, bodyEnd: durationSec)]
}
var chunks: [PlannedChunk] = []
var start = 0.0
var bodyStart = 0.0
var index = 0
while start < durationSec - 0.001 {
let end = min(start + chunkSeconds, durationSec)
chunks.append(PlannedChunk(index: index, start: start, end: end))
start = end
while bodyStart < durationSec - 0.001 {
let bodyEnd = min(bodyStart + chunkSeconds, durationSec)
chunks.append(PlannedChunk(
index: index,
start: max(0, bodyStart - overlapSeconds),
end: min(durationSec, bodyEnd + overlapSeconds),
bodyStart: bodyStart, bodyEnd: bodyEnd))
bodyStart = bodyEnd
index += 1
}
return chunks