Commit Graph

  • 85146e9947 chore: bump version to 1.15.10 main Hawk Lee 2026-02-24 23:13:08 +08:00
  • 8272f7b9a0 fix(stitcher): 修复 VAD onset 选错段 + padding 突破 hard limit Hawk Lee 2026-02-24 23:10:05 +08:00
  • 0534833869 feat(stitcher): 多信号融合边界优化 + 间隙感知 padding Hawk Lee 2026-02-24 22:57:23 +08:00
  • 89d091ca08 fix(stitcher): 重写硬性上界逻辑,直接对下一句做 VAD/Energy 精修获取真实起始位置 (v1.15.8) Hawk Lee 2026-02-24 12:24:44 +08:00
  • ec383c450b fix(stitcher): 修复 FA+VAD 硬性上界导致 cut_end < cut_start 的 bug,增加调试日志 (v1.15.7) Hawk Lee 2026-02-24 12:17:49 +08:00
  • a5914d8ee7 fix(stitcher): clamp cut_end to next segment's earliest start boundary Hawk Lee 2026-02-24 11:46:42 +08:00
  • 7f1b9f9761 fix: FA stop consonant cutoff and tail allowance; bump version to 1.15.5 Hawk Lee 2026-02-23 23:55:59 +08:00
  • 91d0ba7de1 fix: preserve digits in FA pinyin conversion (MP4 tail truncation) Hawk Lee 2026-02-22 19:29:20 +08:00
  • b0ddef0fb6 fix: role detection regex excludes sentence punctuation (v1.15.3) Hawk Lee 2026-02-22 14:55:00 +08:00
  • 077e210bf2 fix: Podcast Splitter 说话人输出反转 bug (按名称排序而非出现顺序) Hawk Lee 2026-02-22 13:39:14 +08:00
  • f111386663 chore: bump version to 1.15.1 for CI trigger Hawk Lee 2026-02-20 01:33:12 +08:00
  • 0c30d31073 docs: include IndexTTS-2 in project metadata Hawk Lee 2026-02-20 01:22:18 +08:00
  • f73661f655 feat: release version 1.15.0 with IndexTTS-2 inline emotion tags and fixes Hawk Lee 2026-02-20 01:17:02 +08:00
  • 515d749177 fix: remove int16 clipping and use float32 output in infer_v2.py Hawk Lee 2026-02-19 17:09:23 +08:00
  • f74b748eb8 chore: add normalized float saving to test script Hawk Lee 2026-02-19 16:59:58 +08:00
  • ad49343dfc chore: fix patches import order in test script Hawk Lee 2026-02-19 16:36:15 +08:00
  • 32721a6e06 chore: update test script with patches and fp16/fp32 test Hawk Lee 2026-02-19 16:34:24 +08:00
  • a0d3650621 chore: add standalone IndexTTS test script Hawk Lee 2026-02-19 16:26:56 +08:00
  • a5249292bd fix: 恢复 kaldifst import (A/B测试确认必须预加载) Hawk Lee 2026-02-19 16:17:15 +08:00
  • 1b33ca9285 chore: 注释掉 kaldifst import 以验证是否必须 (user request) Hawk Lee 2026-02-19 16:01:53 +08:00
  • d8de95f7f3 fix(IndexTTS): preload kaldifst to prevent NeMo hang (revert NeMo changes) Hawk Lee 2026-02-19 15:52:33 +08:00
  • 8bdc5cb7d3 debug: 使用 faulthandler 和 stderr 确保 watchdog 堆栈输出 Hawk Lee 2026-02-19 15:37:20 +08:00
  • 3b71846c94 debug: 增加 NeMo 导入 30秒 watchdog 打印全线程堆栈 Hawk Lee 2026-02-19 15:15:59 +08:00
  • a707f3bf3c revert: 恢复 aiia_generate_segments.py 到 8b556e8 (完全回滚 NeMo 尝试) Hawk Lee 2026-02-19 15:09:06 +08:00
  • 27ed02e77b fix: 将 transformers 补丁改为临时性的 context manager,执行完立刻恢复 Hawk Lee 2026-02-19 15:06:37 +08:00
  • 00d2b2fa7e fix: NeMo 导入时设置 HF_HUB_OFFLINE=1 防止 w2v-bert-2.0 网络请求卡死 Hawk Lee 2026-02-19 14:55:40 +08:00
  • 9f046e37d4 fix: 直接从 sortformer_diar_models 导入,绕过 aed_multitask_models 卡死 Hawk Lee 2026-02-19 14:44:30 +08:00
  • c8dcc4d9e2 debug: import hook 追踪 NeMo 导入卡死 Hawk Lee 2026-02-19 14:34:41 +08:00
  • dc31f01b0b debug: 检查 modelscope 和 from_pretrained 状态 Hawk Lee 2026-02-19 14:03:01 +08:00
  • a6ee5569f3 debug: 细化 NeMo 导入诊断 Hawk Lee 2026-02-19 13:54:11 +08:00
  • cf705edaa6 debug: 临时诊断 NeMo restore_from 卡住位置 Hawk Lee 2026-02-19 13:45:30 +08:00
  • 29d64b64d3 fix: 删除 from_pretrained 保存/恢复 hack,修复 classmethod 描述符损坏 Hawk Lee 2026-02-19 13:37:11 +08:00
  • 133daf8eee debug: 临时诊断 dispatch_model 空 device_map Hawk Lee 2026-02-19 13:19:47 +08:00
  • 8119b6cb78 fix: 补丁改为永久应用,不再还原 Hawk Lee 2026-02-19 12:57:08 +08:00
  • f7e96d37ce fix: 用 transformers 替代 modelscope 导入,从根源消除副作用 Hawk Lee 2026-02-19 12:51:36 +08:00
  • f7a549022c fix: VibeVoice device_map=auto 导致 accelerate IndexError Hawk Lee 2026-02-19 12:45:03 +08:00
  • 8b3b73c2ed fix: QwenEmotion device_map=auto 与 modelscope 不兼容导致 IndexError Hawk Lee 2026-02-19 12:39:41 +08:00
  • 64a05e4b44 fix: 修复 context manager 双 yield 和 from_pretrained 恢复逻辑 Hawk Lee 2026-02-19 12:35:11 +08:00
  • 579da6b3e4 fix: 启用所有节点模块导入 Hawk Lee 2026-02-19 12:08:47 +08:00
  • 1cb441eaa9 fix: 修复 IndexTTS 导入导致 NeMo 分段节点卡死的问题 Hawk Lee 2026-02-19 12:05:07 +08:00
  • 1246441500 v1.14.3: Podcast Stitcher - use VAD to cap FA cut_end, prevent tail extending into silence Hawk Lee 2026-02-18 00:04:09 +08:00
  • 8b556e8435 v1.14.2: Fix NeMo diarization for PyTorch 2.10+, add JSON Extractor/Builder nodes, Qwen3-TTS voice presets & robustness fixes Hawk Lee 2026-02-17 23:10:27 +08:00
  • fb0c84bfae v1.14.1: Add silence gap between TTS segments & fix Qwen3 TTS stitching Hawk Lee 2026-02-17 16:13:08 +08:00
  • f95f2a9832 v1.14.0: Fix CosyVoice transformers 4.57+ compatibility & emotion segment crossfade Hawk Lee 2026-02-17 15:49:12 +08:00
  • e21a5c9eba fix: restore file permissions to 755 (reverts eff537d mode change) Hawk Lee 2026-02-16 12:15:32 +08:00
  • e27987cad6 fix: add llm.base.pt fallback for V3 model switching Hawk Lee 2026-02-16 11:49:50 +08:00
  • 0435d5c932 fix: restore CosyVoice Chinese UI defaults and revert V3 Base special case Hawk Lee 2026-02-16 11:38:17 +08:00
  • 012c4ad7e9 fix: revert V3 Base to inference_instruct2 with minimal instruct Hawk Lee 2026-02-16 11:28:35 +08:00
  • 6a94b32b1d fix: route CosyVoice V3 Base models to zero-shot path Hawk Lee 2026-02-16 11:17:06 +08:00
  • 23d9b87ab0 chore: bump version to 1.13.0 Hawk Lee 2026-02-16 11:06:23 +08:00
  • 40bc5627cb feat: add AIIA Text Splitter node (v1.13.0) Hawk Lee 2026-02-16 11:03:31 +08:00
  • 449183c800 fix: VoiceDesign uses UI design description with priority Hawk Lee 2026-02-16 10:43:00 +08:00
  • 3ef9fd4601 docs: clarify Qwen3 model type emotion support and UI priority Hawk Lee 2026-02-16 10:35:47 +08:00
  • e01e7c3eff fix: UI emotion dropdown takes priority over inline tags Hawk Lee 2026-02-16 10:31:43 +08:00
  • cf5c7aa78c fix: extract and strip inline emotion tags in standalone Qwen3 TTS Hawk Lee 2026-02-16 10:30:51 +08:00
  • 87077453f3 fix: include emotion in Dialogue TTS Qwen3 batch grouping key Hawk Lee 2026-02-16 10:27:21 +08:00
  • 1f76fae615 docs: clarify VibeVoice auto-strips emotion tags Hawk Lee 2026-02-16 10:15:47 +08:00
  • 8d374cbf0c docs: promote Emotion Annotator to top-level section 4.2 Hawk Lee 2026-02-16 10:14:28 +08:00
  • 7f1537b308 fix: strip emotion tags in VibeVoice Realtime node too Hawk Lee 2026-02-16 10:12:44 +08:00
  • 75f11ce028 fix: strip emotion tags before VibeVoice generation Hawk Lee 2026-02-16 10:11:05 +08:00
  • 8dae239185 feat: Splitter embeds emotion tags into output text Hawk Lee 2026-02-16 10:09:25 +08:00
  • 27131510b9 docs: add AIIA Emotion Annotator documentation to README Hawk Lee 2026-02-16 10:02:26 +08:00
  • aea2e9b16a fix: split Qwen3 batches on emotion change Hawk Lee 2026-02-16 01:52:18 +08:00
  • 69a2d799a1 fix: normalize line index to int and show neutral annotations Hawk Lee 2026-02-16 01:44:31 +08:00
  • dc31876404 fix: add User-Agent header to bypass Cloudflare 1010 block Hawk Lee 2026-02-16 01:41:06 +08:00
  • 19ea3ec847 fix: add IS_CHANGED to prevent ComfyUI caching Hawk Lee 2026-02-16 01:39:30 +08:00
  • 68d71bb728 fix: complete EchoMimicV3 dist stubs for single-GPU inference Hawk Lee 2026-02-16 01:37:15 +08:00
  • e6e17aec81 fix: add proxy_url input + EchoMimicV3 dist stub Hawk Lee 2026-02-16 01:34:21 +08:00
  • 7ceb05437d fix: read proxy settings from ~/run.sh when env vars not set Hawk Lee 2026-02-16 01:32:28 +08:00
  • 797c477889 docs: mark LLM emotion annotation as completed Hawk Lee 2026-02-16 01:25:19 +08:00
  • a74a8b20bb feat: add AIIA Emotion Annotator node (LLM-driven) Hawk Lee 2026-02-16 01:24:36 +08:00
  • 4a7132b82f docs: add manual download instructions for MMS FA model Hawk Lee 2026-02-16 00:40:18 +08:00
  • d7476f4d42 feat(fa): auto-copy downloaded model to models/mms_fa/ for centralized management Hawk Lee 2026-02-16 00:39:21 +08:00
  • 1c25316f0d release: v1.12.0 - MMS Forced Alignment integration Hawk Lee 2026-02-16 00:35:12 +08:00
  • 193164f91a fix(fa): add floor guarantee to cut_end clamp - never shorter than FA end Hawk Lee 2026-02-16 00:21:55 +08:00
  • f0b741a6ab fix(fa): clamp cut_end to prevent overlap with next sentence's FA start Hawk Lee 2026-02-16 00:19:39 +08:00
  • c354adbca8 feat(log): show actual hybrid cut points (USED=) in cross-validation log Hawk Lee 2026-02-16 00:13:15 +08:00
  • 0fe63bfa65 fix(fa): hybrid FA start + energy valley end to preserve tail resonance Hawk Lee 2026-02-16 00:05:25 +08:00
  • ab932a030f fix(fa): lowercase all pinyin output and strip non-tokenizable chars Hawk Lee 2026-02-15 23:53:41 +08:00
  • fece4c171a feat(stitcher): integrate MMS Forced Alignment with cross-validation Hawk Lee 2026-02-15 23:43:08 +08:00
  • 708c7d8117 docs: add Splitter/ASR/Stitcher nodes to README Hawk Lee 2026-02-15 23:24:50 +08:00
  • 56fcaef331 feat(stitcher): integrate Silero VAD for precise speech boundary detection Hawk Lee 2026-02-15 23:16:34 +08:00
  • 8a2dc5c53d feat(stitcher): smoothed energy envelope + balanced cut boundaries Hawk Lee 2026-02-15 22:11:03 +08:00
  • 8b351485ca chore: add TODO.md with LLM emotion tagging idea Hawk Lee 2026-02-15 22:05:05 +08:00
  • 210b8d5396 fix(podcast): ensure TTS produces natural sentence-ending pauses Hawk Lee 2026-02-15 21:37:50 +08:00
  • f84c0b2552 fix(stitcher): asymmetric cut boundaries to prevent tail bleed Hawk Lee 2026-02-15 21:19:30 +08:00
  • 308fc8cb1b fix(stitcher): prevent cut_end from bleeding into next sentence Hawk Lee 2026-02-15 20:27:23 +08:00
  • 63d05a8c5e feat(vibevoice): add seed parameter for reproducible speech generation Hawk Lee 2026-02-15 20:20:51 +08:00
  • 96fce4e28d fix(stitcher): prevent repeated audio from padding-induced segment overlap Hawk Lee 2026-02-15 20:05:45 +08:00
  • e33c0b37ea fix(example): correct AudioPostProcess node name to AIIA_Audio_PostProcess Hawk Lee 2026-02-15 19:55:48 +08:00
  • 29e29a2393 fix(example): correct SubtitleGen node name to AIIA_Subtitle_Gen Hawk Lee 2026-02-15 19:54:17 +08:00
  • ed329bc23d fix(stitcher): improve transition naturalness with cosine fade and noise floor Hawk Lee 2026-02-15 19:51:52 +08:00
  • 4ca23f2d9d fix(stitcher): directional cut points to prevent speech onset clipping Hawk Lee 2026-02-15 19:20:01 +08:00
  • e91c905358 fix(browser): restore features lost in eff537d overwrite Hawk Lee 2026-02-15 18:18:23 +08:00
  • 536410b359 feat: add 4 workflow examples + fix pix_fmt regression Hawk Lee 2026-02-15 18:13:46 +08:00
  • d6d6e76a3d docs: add cleanup_frames documentation and v1.11.1 changelog to README Hawk Lee 2026-02-15 17:53:19 +08:00
  • 98cff2b74e docs: sync version to 1.11.1, add cleanup_frames documentation and changelog entry Hawk Lee 2026-02-15 17:51:58 +08:00
  • c1b5f83cb9 feat(safety): add .aiia_temp marker file to prevent accidental deletion of user frame directories Hawk Lee 2026-02-15 17:46:13 +08:00
  • f02eab2669 feat(video): add cleanup_frames option to delete input frames directory after successful merge Hawk Lee 2026-02-15 17:43:11 +08:00
  • 91adc48bf2 fix: unify fps parameter to FLOAT across all nodes; bump to v1.8.3 Hawk Lee 2026-02-15 17:38:15 +08:00