Commit Graph

9 Commits

Author SHA1 Message Date
simonkoson 1c3963d17c doco: v2说话人分段模式 — ASR说话人分离+大block拆分+三维动画解说识别
- asr_adapter: 新增roleType=1说话人分离参数,新增parse_order_result_with_speaker(),write_asr_result自动输出asr_v2_timed_spk.txt

- fusion_align: 新增speaker-aware alignment v2流程(_annotate_b_lines_with_speakers区间匹配、_detect_speaker_blocks、SYSTEM_PROMPT_SPEAKER_ALIGN大block拆分prompt、_build_broadcast_segments支持block内多段拆分)

- cli: 兼容v1/v2 stats字典

- 新增convert_to_md.py(20期融合A稿docx转md+YAML frontmatter)

- backup_before_spk/: 修改前代码备份
2026-06-24 16:26:05 +08:00
simonkoson a87f453326 doco: ep002全流程完工(P1→C4) + CLAUDE.md收摊更新
ep002(潜艇仿生)全部产物入库:B稿v2(733行)/ASR(411句)/融合B稿/
融合A稿.docx/c4_alignment.csv等。fusion_align.py修复align_batch
崩溃bug。CLAUDE.md更新ep002完工状态、新增分段偏差根因和MiMo批次
失败率两条关键决策。

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
2026-06-23 08:15:41 +08:00
simonkoson 010128b123 doco: 更新CLAUDE.md — ep002进度至C3完成/C4跑中
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
2026-06-22 14:37:52 +08:00
simonkoson e213feb814 doco: 更新CLAUDE.md — ep002进度至C2、LLM切MiMo、Ollama并发配通
状态栏/已完成/待办/关键决策/交接备注全部增量更新。

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
2026-06-22 12:59:48 +08:00
simonkoson 8df9c798e7 doco: ep003 全流程跑通(P1→C4)+ OCR并发版 + compose出稿改名 2026-06-18 20:06:45 +08:00
simonkoson 10b0ab757b doco: C4收口 + A稿解析器升级(LLM骨架) + ep002/003/004骨架
- 新增 fusion_align.py(C4 compose + skeleton 提取/校验/预览)与 doco skeleton 命令
- ep001 C4 产物入库: 融合A稿.docx / 融合B稿.txt / c4_alignment.csv / fusion_review.csv
- ep002/003/004: A稿 + 分段骨架(json+预览), 上游待跑
- .gitignore 增 .c3_cache/.c4_cache (可重生中间缓存不入库)
- 更新 doco/CLAUDE.md 交接备注
2026-06-18 14:36:39 +08:00
simonkoson 60f22e6660 P3-C3: 交叉复审 fusion_review.py + doco fuse 命令,产融合B稿743行+fusion_review留痕,更新CLAUDE.md交接C4 2026-06-17 13:27:41 +08:00
simonkoson 0a4ab881bc doco: 更新 CLAUDE.md — C2 推送后解除 C3 卡点(B稿v2 已就位)
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
2026-06-17 10:23:18 +08:00
simonkoson 1372600d71 doco P3-C2: 讯飞ASR适配层接入 + 修复 doco/programs 忽略规则
- C2: asr_adapter 并入讯飞上传/轮询/解析,复用 extract_audio 分离音频,
  新增 doco asr 命令读 C1 热词;真转写 ep001 出 310 句带时间戳
- 补提交 C1 代码(llm.py / term_extract.py)与累积词典
- .gitignore: 挡 doco/programs 下 png/wav/mp4 及中间帧目录、本地 settings

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
2026-06-17 10:19:08 +08:00