Skip to content

wanghai-tech / hikoon-ACR

Go to a project
Toggle navigation
Toggle navigation pinning
Commit 528cc473 528cc473ba8e53a0df86001f564f101a2fc126a1 by cnb.bofCdSsphPA
Options

Turn pgvector planning into repo-native ingestion templates 

Constraint: The user needs concrete downstream data handling guidance now, and future vector retrieval work should not start from abstract docs alone
Rejected: Leave pgvector support at prose-only guidance | Delays integration by forcing later sessions to reinvent schema and export bridges
Confidence: high
Scope-risk: narrow
Directive: Keep schema/export templates aligned with actual manifest semantics before adding live database loaders
Tested: /usr/local/miniconda3/bin/python -m py_compile acr-engine/scripts/export_manifest_to_pgvector_json.py; /usr/local/miniconda3/bin/python acr-engine/scripts/export_manifest_to_pgvector_json.py --data acr-engine/data/synthetic_v2 --split test --source-dataset synthetic_v2 --output acr-engine/reports/pgvector_manifest_export_test.json
Not-tested: Live PostgreSQL/pgvector ingestion remains pending a real database target
1 parent d6d67893
Inline Side-by-side
Showing 5 changed files with 756 additions and 0 deletions
1 {
2 "songs": [
3 {
4 "song_id": "song_0000",
5 "title": "song_0000",
6 "artist": null,
7 "version_id": null,
8 "source_dataset": "synthetic_v2",
9 "license": null
10 },
11 {
12 "song_id": "song_0001",
13 "title": "song_0001",
14 "artist": null,
15 "version_id": null,
16 "source_dataset": "synthetic_v2",
17 "license": null
18 },
19 {
20 "song_id": "song_0002",
21 "title": "song_0002",
22 "artist": null,
23 "version_id": null,
24 "source_dataset": "synthetic_v2",
25 "license": null
26 },
27 {
28 "song_id": "song_0003",
29 "title": "song_0003",
30 "artist": null,
31 "version_id": null,
32 "source_dataset": "synthetic_v2",
33 "license": null
34 },
35 {
36 "song_id": "song_0004",
37 "title": "song_0004",
38 "artist": null,
39 "version_id": null,
40 "source_dataset": "synthetic_v2",
41 "license": null
42 },
43 {
44 "song_id": "song_0005",
45 "title": "song_0005",
46 "artist": null,
47 "version_id": null,
48 "source_dataset": "synthetic_v2",
49 "license": null
50 },
51 {
52 "song_id": "song_0006",
53 "title": "song_0006",
54 "artist": null,
55 "version_id": null,
56 "source_dataset": "synthetic_v2",
57 "license": null
58 },
59 {
60 "song_id": "song_0007",
61 "title": "song_0007",
62 "artist": null,
63 "version_id": null,
64 "source_dataset": "synthetic_v2",
65 "license": null
66 },
67 {
68 "song_id": "song_0008",
69 "title": "song_0008",
70 "artist": null,
71 "version_id": null,
72 "source_dataset": "synthetic_v2",
73 "license": null
74 },
75 {
76 "song_id": "song_0009",
77 "title": "song_0009",
78 "artist": null,
79 "version_id": null,
80 "source_dataset": "synthetic_v2",
81 "license": null
82 },
83 {
84 "song_id": "song_0010",
85 "title": "song_0010",
86 "artist": null,
87 "version_id": null,
88 "source_dataset": "synthetic_v2",
89 "license": null
90 },
91 {
92 "song_id": "song_0011",
93 "title": "song_0011",
94 "artist": null,
95 "version_id": null,
96 "source_dataset": "synthetic_v2",
97 "license": null
98 },
99 {
100 "song_id": "song_0012",
101 "title": "song_0012",
102 "artist": null,
103 "version_id": null,
104 "source_dataset": "synthetic_v2",
105 "license": null
106 },
107 {
108 "song_id": "song_0013",
109 "title": "song_0013",
110 "artist": null,
111 "version_id": null,
112 "source_dataset": "synthetic_v2",
113 "license": null
114 },
115 {
116 "song_id": "song_0014",
117 "title": "song_0014",
118 "artist": null,
119 "version_id": null,
120 "source_dataset": "synthetic_v2",
121 "license": null
122 },
123 {
124 "song_id": "song_0015",
125 "title": "song_0015",
126 "artist": null,
127 "version_id": null,
128 "source_dataset": "synthetic_v2",
129 "license": null
130 },
131 {
132 "song_id": "song_0016",
133 "title": "song_0016",
134 "artist": null,
135 "version_id": null,
136 "source_dataset": "synthetic_v2",
137 "license": null
138 },
139 {
140 "song_id": "song_0017",
141 "title": "song_0017",
142 "artist": null,
143 "version_id": null,
144 "source_dataset": "synthetic_v2",
145 "license": null
146 },
147 {
148 "song_id": "song_0018",
149 "title": "song_0018",
150 "artist": null,
151 "version_id": null,
152 "source_dataset": "synthetic_v2",
153 "license": null
154 },
155 {
156 "song_id": "song_0019",
157 "title": "song_0019",
158 "artist": null,
159 "version_id": null,
160 "source_dataset": "synthetic_v2",
161 "license": null
162 },
163 {
164 "song_id": "song_0020",
165 "title": "song_0020",
166 "artist": null,
167 "version_id": null,
168 "source_dataset": "synthetic_v2",
169 "license": null
170 },
171 {
172 "song_id": "song_0021",
173 "title": "song_0021",
174 "artist": null,
175 "version_id": null,
176 "source_dataset": "synthetic_v2",
177 "license": null
178 },
179 {
180 "song_id": "song_0022",
181 "title": "song_0022",
182 "artist": null,
183 "version_id": null,
184 "source_dataset": "synthetic_v2",
185 "license": null
186 },
187 {
188 "song_id": "song_0023",
189 "title": "song_0023",
190 "artist": null,
191 "version_id": null,
192 "source_dataset": "synthetic_v2",
193 "license": null
194 }
195 ],
196 "references": [
197 {
198 "song_id": "song_0000",
199 "audio_uri": "songs/song_0000.wav",
200 "duration_sec": 15.0,
201 "sample_rate": 16000
202 },
203 {
204 "song_id": "song_0001",
205 "audio_uri": "songs/song_0001.wav",
206 "duration_sec": 15.0,
207 "sample_rate": 16000
208 },
209 {
210 "song_id": "song_0002",
211 "audio_uri": "songs/song_0002.wav",
212 "duration_sec": 15.0,
213 "sample_rate": 16000
214 },
215 {
216 "song_id": "song_0003",
217 "audio_uri": "songs/song_0003.wav",
218 "duration_sec": 15.0,
219 "sample_rate": 16000
220 },
221 {
222 "song_id": "song_0004",
223 "audio_uri": "songs/song_0004.wav",
224 "duration_sec": 15.0,
225 "sample_rate": 16000
226 },
227 {
228 "song_id": "song_0005",
229 "audio_uri": "songs/song_0005.wav",
230 "duration_sec": 15.0,
231 "sample_rate": 16000
232 },
233 {
234 "song_id": "song_0006",
235 "audio_uri": "songs/song_0006.wav",
236 "duration_sec": 15.0,
237 "sample_rate": 16000
238 },
239 {
240 "song_id": "song_0007",
241 "audio_uri": "songs/song_0007.wav",
242 "duration_sec": 15.0,
243 "sample_rate": 16000
244 },
245 {
246 "song_id": "song_0008",
247 "audio_uri": "songs/song_0008.wav",
248 "duration_sec": 15.0,
249 "sample_rate": 16000
250 },
251 {
252 "song_id": "song_0009",
253 "audio_uri": "songs/song_0009.wav",
254 "duration_sec": 15.0,
255 "sample_rate": 16000
256 },
257 {
258 "song_id": "song_0010",
259 "audio_uri": "songs/song_0010.wav",
260 "duration_sec": 15.0,
261 "sample_rate": 16000
262 },
263 {
264 "song_id": "song_0011",
265 "audio_uri": "songs/song_0011.wav",
266 "duration_sec": 15.0,
267 "sample_rate": 16000
268 },
269 {
270 "song_id": "song_0012",
271 "audio_uri": "songs/song_0012.wav",
272 "duration_sec": 15.0,
273 "sample_rate": 16000
274 },
275 {
276 "song_id": "song_0013",
277 "audio_uri": "songs/song_0013.wav",
278 "duration_sec": 15.0,
279 "sample_rate": 16000
280 },
281 {
282 "song_id": "song_0014",
283 "audio_uri": "songs/song_0014.wav",
284 "duration_sec": 15.0,
285 "sample_rate": 16000
286 },
287 {
288 "song_id": "song_0015",
289 "audio_uri": "songs/song_0015.wav",
290 "duration_sec": 15.0,
291 "sample_rate": 16000
292 },
293 {
294 "song_id": "song_0016",
295 "audio_uri": "songs/song_0016.wav",
296 "duration_sec": 15.0,
297 "sample_rate": 16000
298 },
299 {
300 "song_id": "song_0017",
301 "audio_uri": "songs/song_0017.wav",
302 "duration_sec": 15.0,
303 "sample_rate": 16000
304 },
305 {
306 "song_id": "song_0018",
307 "audio_uri": "songs/song_0018.wav",
308 "duration_sec": 15.0,
309 "sample_rate": 16000
310 },
311 {
312 "song_id": "song_0019",
313 "audio_uri": "songs/song_0019.wav",
314 "duration_sec": 15.0,
315 "sample_rate": 16000
316 },
317 {
318 "song_id": "song_0020",
319 "audio_uri": "songs/song_0020.wav",
320 "duration_sec": 15.0,
321 "sample_rate": 16000
322 },
323 {
324 "song_id": "song_0021",
325 "audio_uri": "songs/song_0021.wav",
326 "duration_sec": 15.0,
327 "sample_rate": 16000
328 },
329 {
330 "song_id": "song_0022",
331 "audio_uri": "songs/song_0022.wav",
332 "duration_sec": 15.0,
333 "sample_rate": 16000
334 },
335 {
336 "song_id": "song_0023",
337 "audio_uri": "songs/song_0023.wav",
338 "duration_sec": 15.0,
339 "sample_rate": 16000
340 }
341 ],
342 "segments": [
343 {
344 "song_id": "song_0020",
345 "audio_uri": "segments/song_0020_seg_00.wav",
346 "offset_sec": 4.349828784349853,
347 "duration_sec": 5.0,
348 "split": "test",
349 "type": "clean",
350 "segment_type": "mid",
351 "source_dataset": "synthetic_v2"
352 },
353 {
354 "song_id": "song_0020",
355 "audio_uri": "segments/song_0020_seg_01.wav",
356 "offset_sec": 9.642182747327407,
357 "duration_sec": 5.0,
358 "split": "test",
359 "type": "clean",
360 "segment_type": "mid",
361 "source_dataset": "synthetic_v2"
362 },
363 {
364 "song_id": "song_0020",
365 "audio_uri": "segments/song_0020_seg_02_augmented.wav",
366 "offset_sec": 2.367717347418965,
367 "duration_sec": 5.0,
368 "split": "test",
369 "type": "augmented",
370 "segment_type": "intro",
371 "source_dataset": "synthetic_v2"
372 },
373 {
374 "song_id": "song_0020",
375 "audio_uri": "segments/song_0020_seg_03_humming_like.wav",
376 "offset_sec": 3.180577192661006,
377 "duration_sec": 5.0,
378 "split": "test",
379 "type": "humming_like",
380 "segment_type": "mid",
381 "source_dataset": "synthetic_v2"
382 },
383 {
384 "song_id": "song_0020",
385 "audio_uri": "segments/song_0020_seg_04_confused.wav",
386 "offset_sec": 4.660551124366617,
387 "duration_sec": 5.0,
388 "split": "test",
389 "type": "confused",
390 "segment_type": "mid",
391 "source_dataset": "synthetic_v2"
392 },
393 {
394 "song_id": "song_0021",
395 "audio_uri": "segments/song_0021_seg_00.wav",
396 "offset_sec": 5.631088908640184,
397 "duration_sec": 5.0,
398 "split": "test",
399 "type": "clean",
400 "segment_type": "mid",
401 "source_dataset": "synthetic_v2"
402 },
403 {
404 "song_id": "song_0021",
405 "audio_uri": "segments/song_0021_seg_01.wav",
406 "offset_sec": 1.8823366490525628,
407 "duration_sec": 5.0,
408 "split": "test",
409 "type": "clean",
410 "segment_type": "intro",
411 "source_dataset": "synthetic_v2"
412 },
413 {
414 "song_id": "song_0021",
415 "audio_uri": "segments/song_0021_seg_02_augmented.wav",
416 "offset_sec": 9.88006210404643,
417 "duration_sec": 5.0,
418 "split": "test",
419 "type": "augmented",
420 "segment_type": "mid",
421 "source_dataset": "synthetic_v2"
422 },
423 {
424 "song_id": "song_0021",
425 "audio_uri": "segments/song_0021_seg_03_humming_like.wav",
426 "offset_sec": 0.9025737685090285,
427 "duration_sec": 5.0,
428 "split": "test",
429 "type": "humming_like",
430 "segment_type": "intro",
431 "source_dataset": "synthetic_v2"
432 },
433 {
434 "song_id": "song_0021",
435 "audio_uri": "segments/song_0021_seg_04_confused.wav",
436 "offset_sec": 1.3048954561918258,
437 "duration_sec": 5.0,
438 "split": "test",
439 "type": "confused",
440 "segment_type": "intro",
441 "source_dataset": "synthetic_v2"
442 },
443 {
444 "song_id": "song_0022",
445 "audio_uri": "segments/song_0022_seg_00.wav",
446 "offset_sec": 3.9746734850812295,
447 "duration_sec": 5.0,
448 "split": "test",
449 "type": "clean",
450 "segment_type": "mid",
451 "source_dataset": "synthetic_v2"
452 },
453 {
454 "song_id": "song_0022",
455 "audio_uri": "segments/song_0022_seg_01.wav",
456 "offset_sec": 4.890968121206573,
457 "duration_sec": 5.0,
458 "split": "test",
459 "type": "clean",
460 "segment_type": "mid",
461 "source_dataset": "synthetic_v2"
462 },
463 {
464 "song_id": "song_0022",
465 "audio_uri": "segments/song_0022_seg_02_augmented.wav",
466 "offset_sec": 6.610400547460049,
467 "duration_sec": 5.0,
468 "split": "test",
469 "type": "augmented",
470 "segment_type": "mid",
471 "source_dataset": "synthetic_v2"
472 },
473 {
474 "song_id": "song_0022",
475 "audio_uri": "segments/song_0022_seg_03_humming_like.wav",
476 "offset_sec": 2.6329596668288424,
477 "duration_sec": 5.0,
478 "split": "test",
479 "type": "humming_like",
480 "segment_type": "intro",
481 "source_dataset": "synthetic_v2"
482 },
483 {
484 "song_id": "song_0022",
485 "audio_uri": "segments/song_0022_seg_04_confused.wav",
486 "offset_sec": 0.8570731183991709,
487 "duration_sec": 5.0,
488 "split": "test",
489 "type": "confused",
490 "segment_type": "intro",
491 "source_dataset": "synthetic_v2"
492 },
493 {
494 "song_id": "song_0023",
495 "audio_uri": "segments/song_0023_seg_00.wav",
496 "offset_sec": 4.461034326075292,
497 "duration_sec": 5.0,
498 "split": "test",
499 "type": "clean",
500 "segment_type": "mid",
501 "source_dataset": "synthetic_v2"
502 },
503 {
504 "song_id": "song_0023",
505 "audio_uri": "segments/song_0023_seg_01.wav",
506 "offset_sec": 9.605203782802876,
507 "duration_sec": 5.0,
508 "split": "test",
509 "type": "clean",
510 "segment_type": "mid",
511 "source_dataset": "synthetic_v2"
512 },
513 {
514 "song_id": "song_0023",
515 "audio_uri": "segments/song_0023_seg_02_augmented.wav",
516 "offset_sec": 4.7458228906154805,
517 "duration_sec": 5.0,
518 "split": "test",
519 "type": "augmented",
520 "segment_type": "mid",
521 "source_dataset": "synthetic_v2"
522 },
523 {
524 "song_id": "song_0023",
525 "audio_uri": "segments/song_0023_seg_03_humming_like.wav",
526 "offset_sec": 8.308702013555955,
527 "duration_sec": 5.0,
528 "split": "test",
529 "type": "humming_like",
530 "segment_type": "mid",
531 "source_dataset": "synthetic_v2"
532 },
533 {
534 "song_id": "song_0023",
535 "audio_uri": "segments/song_0023_seg_04_confused.wav",
536 "offset_sec": 2.213510770155481,
537 "duration_sec": 5.0,
538 "split": "test",
539 "type": "confused",
540 "segment_type": "intro",
541 "source_dataset": "synthetic_v2"
542 }
543 ]
544 }
...\ No newline at end of file ...\ No newline at end of file
1 #!/usr/bin/env python3
2 """Export project manifests into a pgvector-friendly JSON payload.
3
4 This does not require PostgreSQL at runtime. It prepares normalized rows so a
5 future loader can bulk ingest them into Postgres/pgvector safely.
6 """
7
8 from __future__ import annotations
9
10 import argparse
11 import json
12 from pathlib import Path
13
14
15 def load_json(path: Path):
16 return json.loads(path.read_text())
17
18
19 def main():
20 parser = argparse.ArgumentParser()
21 parser.add_argument("--data", required=True, help="manifest directory")
22 parser.add_argument("--output", required=True)
23 parser.add_argument("--split", default="train")
24 parser.add_argument("--source-dataset", default="unknown")
25 args = parser.parse_args()
26
27 data_dir = Path(args.data)
28 catalog = load_json(data_dir / "catalog.json")
29 split_rows = load_json(data_dir / f"{args.split}.json")
30
31 songs = {}
32 references = []
33 segments = []
34
35 for row in catalog:
36 song_id = row["song_id"]
37 songs.setdefault(song_id, {
38 "song_id": song_id,
39 "title": song_id,
40 "artist": None,
41 "version_id": None,
42 "source_dataset": row.get("source_dataset", args.source_dataset),
43 "license": None,
44 })
45 if row.get("type") == "reference":
46 references.append({
47 "song_id": song_id,
48 "audio_uri": row["audio_path"],
49 "duration_sec": row["duration"],
50 "sample_rate": 16000,
51 })
52
53 for row in split_rows:
54 if row.get("type") == "reference":
55 continue
56 song_id = row["song_id"]
57 songs.setdefault(song_id, {
58 "song_id": song_id,
59 "title": song_id,
60 "artist": None,
61 "version_id": None,
62 "source_dataset": row.get("source_dataset", args.source_dataset),
63 "license": None,
64 })
65 segments.append({
66 "song_id": song_id,
67 "audio_uri": row["audio_path"],
68 "offset_sec": row.get("offset", 0.0),
69 "duration_sec": row["duration"],
70 "split": args.split,
71 "type": row.get("type", "unknown"),
72 "segment_type": row.get("segment_type"),
73 "source_dataset": row.get("source_dataset", args.source_dataset),
74 })
75
76 payload = {
77 "songs": list(songs.values()),
78 "references": references,
79 "segments": segments,
80 }
81
82 out = Path(args.output)
83 out.parent.mkdir(parents=True, exist_ok=True)
84 out.write_text(json.dumps(payload, indent=2, ensure_ascii=False))
85 print(json.dumps({
86 "status": "ok",
87 "output": str(out.resolve()),
88 "songs": len(payload["songs"]),
89 "references": len(payload["references"]),
90 "segments": len(payload["segments"]),
91 }, indent=2, ensure_ascii=False))
92
93
94 if __name__ == "__main__":
95 main()
1 CREATE EXTENSION IF NOT EXISTS vector;
2
3 CREATE TABLE IF NOT EXISTS songs (
4 song_id TEXT PRIMARY KEY,
5 title TEXT,
6 artist TEXT,
7 version_id TEXT,
8 source_dataset TEXT,
9 license TEXT,
10 created_at TIMESTAMPTZ DEFAULT NOW()
11 );
12
13 CREATE TABLE IF NOT EXISTS references (
14 reference_id BIGSERIAL PRIMARY KEY,
15 song_id TEXT NOT NULL REFERENCES songs(song_id) ON DELETE CASCADE,
16 audio_uri TEXT NOT NULL,
17 duration_sec DOUBLE PRECISION NOT NULL,
18 sample_rate INTEGER DEFAULT 16000,
19 created_at TIMESTAMPTZ DEFAULT NOW()
20 );
21
22 CREATE TABLE IF NOT EXISTS segments (
23 segment_id BIGSERIAL PRIMARY KEY,
24 song_id TEXT NOT NULL REFERENCES songs(song_id) ON DELETE CASCADE,
25 audio_uri TEXT NOT NULL,
26 offset_sec DOUBLE PRECISION DEFAULT 0,
27 duration_sec DOUBLE PRECISION NOT NULL,
28 split TEXT,
29 type TEXT NOT NULL,
30 segment_type TEXT,
31 source_dataset TEXT,
32 created_at TIMESTAMPTZ DEFAULT NOW()
33 );
34
35 CREATE TABLE IF NOT EXISTS reference_embeddings (
36 embedding_id BIGSERIAL PRIMARY KEY,
37 reference_id BIGINT NOT NULL REFERENCES references(reference_id) ON DELETE CASCADE,
38 song_id TEXT NOT NULL REFERENCES songs(song_id) ON DELETE CASCADE,
39 embedding vector(192) NOT NULL,
40 model_version TEXT NOT NULL,
41 data_version TEXT,
42 created_at TIMESTAMPTZ DEFAULT NOW()
43 );
44
45 CREATE TABLE IF NOT EXISTS query_embeddings (
46 embedding_id BIGSERIAL PRIMARY KEY,
47 segment_id BIGINT NOT NULL REFERENCES segments(segment_id) ON DELETE CASCADE,
48 song_id TEXT NOT NULL REFERENCES songs(song_id) ON DELETE CASCADE,
49 embedding vector(192) NOT NULL,
50 model_version TEXT NOT NULL,
51 data_version TEXT,
52 created_at TIMESTAMPTZ DEFAULT NOW()
53 );
54
55 CREATE INDEX IF NOT EXISTS idx_segments_song_id ON segments(song_id);
56 CREATE INDEX IF NOT EXISTS idx_references_song_id ON references(song_id);
57 CREATE INDEX IF NOT EXISTS idx_reference_embeddings_song_id ON reference_embeddings(song_id);
58 CREATE INDEX IF NOT EXISTS idx_query_embeddings_song_id ON query_embeddings(song_id);
59
60 CREATE INDEX IF NOT EXISTS idx_reference_embeddings_vector_cosine
61 ON reference_embeddings USING ivfflat (embedding vector_cosine_ops)
62 WITH (lists = 100);
63
64 CREATE INDEX IF NOT EXISTS idx_query_embeddings_vector_cosine
65 ON query_embeddings USING ivfflat (embedding vector_cosine_ops)
66 WITH (lists = 100);
...@@ -234,6 +234,26 @@ ...@@ -234,6 +234,26 @@
234 234
235 235
236 236
237
238 ### Stage: pgvector 落库模板
239
240 完成项：
241 - 新增 [acr-engine/sql/pgvector_schema.sql](../acr-engine/sql/pgvector_schema.sql)
242 - 新增 [acr-engine/scripts/export_manifest_to_pgvector_json.py](../acr-engine/scripts/export_manifest_to_pgvector_json.py)
243 - 在 [docs/training-data-and-pgvector-guide.md](./training-data-and-pgvector-guide.md) 中补充可执行模板说明
244
245 验证结果：
246 - `/usr/local/miniconda3/bin/python -m py_compile scripts/export_manifest_to_pgvector_json.py` 成功
247 - `/usr/local/miniconda3/bin/python scripts/export_manifest_to_pgvector_json.py --data data/synthetic_v2 --split test --source-dataset synthetic_v2 --output reports/pgvector_manifest_export_test.json` 成功
248 - 当前导出结果：
249 - `songs=24`
250 - `references=24`
251 - `segments=20`
252
253 结论：
254 - pgvector 方向现在不仅有概念文档，还有可直接复用的 schema 和 manifest 导出桥接脚本
255 - 后续接 PostgreSQL 时返工成本会显著降低
256
237 ### Stage: FMA 下载自动守护 257 ### Stage: FMA 下载自动守护
238 258
239 完成项： 259 完成项：
......
...@@ -508,6 +508,37 @@ val.json ...@@ -508,6 +508,37 @@ val.json
508 - [dataset-sources-and-licensing.md](./dataset-sources-and-licensing.md) 508 - [dataset-sources-and-licensing.md](./dataset-sources-and-licensing.md)
509 - [session-handoff.md](./session-handoff.md) 509 - [session-handoff.md](./session-handoff.md)
510 510
511
512 ## 12. 可直接落地的 pgvector 模板
513
514 仓库里现在已经补了两个可直接参考的模板：
515
516 - SQL schema: [acr-engine/sql/pgvector_schema.sql](../acr-engine/sql/pgvector_schema.sql)
517 - manifest 导出桥接脚本: [acr-engine/scripts/export_manifest_to_pgvector_json.py](../acr-engine/scripts/export_manifest_to_pgvector_json.py)
518
519 ### 导出示例
520
521 ```bash
522 cd acr-engine
523 /usr/local/miniconda3/bin/python scripts/export_manifest_to_pgvector_json.py \
524 --data data/synthetic_v2 \
525 --split test \
526 --source-dataset synthetic_v2 \
527 --output reports/pgvector_manifest_export_test.json
528 ```
529
530 ### 当前已验证结果
531
532 - `songs=24`
533 - `references=24`
534 - `segments=20`
535
536 这一步还不会直接写 PostgreSQL，作用是：
537
538 1. 先把项目现有 manifest 规范转换成 pgvector-friendly 结构化 JSON
539 2. 后续你们可以再用 bulk insert / COPY / ETL 把这些行落到 PostgreSQL
540 3. embedding 生成后再写入 `vector(192)` 列
541
511 ## Sources 542 ## Sources
512 543
513 - Current code behavior from: 544 - Current code behavior from:
......