[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M7C0Z1TP5DPRQH0CBMQ4MX7F":3,"skill-same:01M7C0Z1TP5DPRQH0CBMQ4MX7F":45},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":16,"categories":17,"keywords":18,"publishedAt":19,"linkOnly":20,"markdownContent":21,"markdownBody":22,"frontMatter":23,"refName":24,"commitSha":25,"skillDirectory":26,"forksCount":27,"repositoryHtmlUrl":28,"repositoryHomepageUrl":6,"ownerHtmlUrl":29,"ownerBlogUrl":6,"ownerDescription":6,"updatedAt":30,"install":31,"resources":44},"01M7C0Z1TP5DPRQH0CBMQ4MX7F","text-to-speech",null,"Practical mastering steps for TTS audio: cleanup, loudness normalization, alignment, and delivery specs.","linyouxue\u002Fskill-repair-benchmark","linyouxue","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F216151464?v=4","01M7BJEVCXKD74KH6FG9CBVSQB","林宥学","01M7BJEVE02NP217V93BKZFC0J",0,"Apache-2.0","https:\u002F\u002Fgithub.com\u002Flinyouxue\u002Fskill-repair-benchmark\u002Fblob\u002Fmain\u002Fevaluation\u002Fresults\u002FSkillOpt-adapted-gpt52-n3L4R2-v25\u002Ftasks\u002Fmultilingual-video-dubbing\u002Frepaired_skill\u002Ftext-to-speech\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\n\nname: text-to-speech\ndescription: \"Practical mastering steps for TTS audio: cleanup, loudness normalization, alignment, and delivery specs.\"\n---\n\n# SKILL: TTS Audio Mastering\n\nThis skill focuses on producing clean, consistent, and delivery-ready TTS audio for video tasks. It covers speech cleanup, loudness normalization, segment boundaries, and export specs.\n\n## 1. TTS Engine & Output Basics\n\nChoose a TTS engine based on deployment constraints and quality needs:\n\n* **Neural offline** (e.g., Kokoro): stable, high quality, no network dependency.\n* **Cloud TTS** (e.g., Edge-TTS \u002F OpenAI TTS): convenient, higher naturalness but network-dependent.\n* **Formant TTS** (e.g., espeak-ng): for prototyping only; often less natural.\n\n**Key rule:** Always confirm the **native sample rate** of the generated audio before resampling for video delivery.\n\n---\n\n## 2. Speech Cleanup (Per Segment)\n\nApply lightweight processing to avoid common artifacts:\n\n* **Rumble\u002FDC removal:** high-pass filter around **20 Hz**\n* **Harshness control:** optional low-pass around **16 kHz** (helps remove digital fizz)\n* **Click\u002Fpop prevention:** short fades at boundaries (e.g., **50 ms** fade-in and fade-out)\n\nRecommended FFmpeg pattern (example):\n\n* Add filters in a single chain, and keep them consistent across segments.\n\n---\n\n## 3. Loudness Normalization\n\nTarget loudness depends on the benchmark\u002Ftask spec. A common target is ITU-R BS.1770 loudness measurement:\n\n* **Integrated loudness:** **-23 LUFS**\n* **True peak:** around **-1.5 dBTP**\n* **LRA:** around **11** (optional)\n\nRecommended workflow:\n\n1. **Measure loudness** using FFmpeg `ebur128` (or equivalent meter).\n2. **Apply normalization** (e.g., `loudnorm`) as the final step after cleanup and timing edits.\n3. If you adjust tempo\u002Fduration after normalization, re-normalize again.\n\n---\n\n## 4. Timing & Segment Boundary Handling\n\nFor multilingual dubbing, treat `\u002Froot\u002Fsegments.srt` as the authoritative placement timeline. Parse SRT timestamps exactly (accept both comma and period millisecond separators), retain integer `start_ms` and `end_ms`, and compute `window_duration_ms = end_ms - start_ms` without rounding. Parse `\u002Froot\u002Fsource_text.srt` and `\u002Froot\u002Freference_target_text.srt` with the same parser and pair cues one-to-one by stable subtitle index\u002Forder; fail rather than silently dropping, duplicating, or inventing a cue when the counts or indices do not match.\n\nSynthesize every paired target cue independently and preserve stable cue order. Measure **raw speech duration before adding silence padding, fades, placement, loudness processing, or resampling**. Fit that speech to its assigned window in this order:\n\n1. Use a conservative, bounded speaking-rate adjustment or time-stretch first, then remeasure. Chain bounded tempo factors when needed rather than applying an extreme single factor.\n2. If the speech is shorter than the window, preserve it and append silence to the window boundary.\n3. Trim only as a last resort if the bounded adjustment still cannot fit the window.\n\nRecord the control actually used for each cue as exactly one of `rate_adjust`, `pad_silence`, or `trim`; do not report a planned control if the rendered result used another one. Keep each fitted cue within its assigned window. Apply short fades only after timing edits, and ensure fades do not change the reported placement boundaries. Resampling, cleanup, normalization, and fades must likewise leave the recorded window start\u002Fend and placement calculations unchanged.\n\nCreate one mono WAV per cue in `\u002Foutputs\u002Ftts_segments\u002F`, named in stable index order (`seg_0.wav`, `seg_1.wav`, ...). `seg_0.wav` is the required first-cue artifact, not the complete program timeline. Store the complete assembled timeline separately (for example, `\u002Foutputs\u002Ftts_segments\u002Fassembled_timeline.wav`) with silence outside cue windows, using the integer millisecond offsets from `segments.srt` rather than concatenating clips. The assembled timeline is the audio to normalize and mux; retain the individual cue files for artifact and duration verification.\n\nFor every cue, report the measured raw\u002Ffitted `tts_duration_sec`, `window_start_sec`, `window_end_sec`, actual rendered `placed_start_sec`, `placed_end_sec`, and `drift_sec = placed_end_sec - window_end_sec`. Verify `abs(placed_start_sec - window_start_sec) \u003C= 0.010` seconds and `abs(drift_sec) \u003C= 0.2` seconds before delivery. Render the final assembled audio at 48000 Hz, mono, and measure\u002Fnormalize it using an ITU-R BS.1770-compatible method after timing edits.\n\n\n\n## 5. Multilingual Input, Voice, and Delivery Contract\n\nRead `\u002Froot\u002Ftarget_language.txt` deterministically: use its first non-empty, trimmed line as the target-language specification, normalize it to a language code for reporting, and do not substitute a guessed language. Select a voice that explicitly supports that target language. Prefer an installed offline-capable neural voice; if the preferred engine or voice is unavailable, use an installed offline fallback such as `espeak-ng` with the closest verified target-language voice and record the fallback choice. Never silently synthesize in the source language.\n\nUse the target text from `\u002Froot\u002Freference_target_text.srt` verbatim after parsing; use `\u002Froot\u002Fsource_text.srt` only for source-text pairing and reporting. Preserve cue text and indices, and fail if any required input is missing or pairing is ambiguous. Inspect the generated WAVs and the separately assembled timeline with media metadata tools before writing outputs.\n\nThe complete normalized assembled timeline, not `\u002Foutputs\u002Ftts_segments\u002Fseg_0.wav` unless there is only one cue, must be used to create `\u002Foutputs\u002Fdubbed.mp4` while preserving the original visual stream. The final audio must be 48000 Hz and mono. Write `\u002Foutputs\u002Freport.json` only after validation, using language codes and including the actual measured loudness, original and final durations, audio properties, every paired cue's source\u002Ftarget text, placement boundaries, drift, measured TTS duration, and its actual `duration_control`. Keep the complete timeline artifact separate from the required first-segment artifact and make this distinction explicit in any internal metadata or report.\n","# SKILL: TTS Audio Mastering\n\nThis skill focuses on producing clean, consistent, and delivery-ready TTS audio for video tasks. It covers speech cleanup, loudness normalization, segment boundaries, and export specs.\n\n## 1. TTS Engine & Output Basics\n\nChoose a TTS engine based on deployment constraints and quality needs:\n\n* **Neural offline** (e.g., Kokoro): stable, high quality, no network dependency.\n* **Cloud TTS** (e.g., Edge-TTS \u002F OpenAI TTS): convenient, higher naturalness but network-dependent.\n* **Formant TTS** (e.g., espeak-ng): for prototyping only; often less natural.\n\n**Key rule:** Always confirm the **native sample rate** of the generated audio before resampling for video delivery.\n\n---\n\n## 2. Speech Cleanup (Per Segment)\n\nApply lightweight processing to avoid common artifacts:\n\n* **Rumble\u002FDC removal:** high-pass filter around **20 Hz**\n* **Harshness control:** optional low-pass around **16 kHz** (helps remove digital fizz)\n* **Click\u002Fpop prevention:** short fades at boundaries (e.g., **50 ms** fade-in and fade-out)\n\nRecommended FFmpeg pattern (example):\n\n* Add filters in a single chain, and keep them consistent across segments.\n\n---\n\n## 3. Loudness Normalization\n\nTarget loudness depends on the benchmark\u002Ftask spec. A common target is ITU-R BS.1770 loudness measurement:\n\n* **Integrated loudness:** **-23 LUFS**\n* **True peak:** around **-1.5 dBTP**\n* **LRA:** around **11** (optional)\n\nRecommended workflow:\n\n1. **Measure loudness** using FFmpeg `ebur128` (or equivalent meter).\n2. **Apply normalization** (e.g., `loudnorm`) as the final step after cleanup and timing edits.\n3. If you adjust tempo\u002Fduration after normalization, re-normalize again.\n\n---\n\n## 4. Timing & Segment Boundary Handling\n\nFor multilingual dubbing, treat `\u002Froot\u002Fsegments.srt` as the authoritative placement timeline. Parse SRT timestamps exactly (accept both comma and period millisecond separators), retain integer `start_ms` and `end_ms`, and compute `window_duration_ms = end_ms - start_ms` without rounding. Parse `\u002Froot\u002Fsource_text.srt` and `\u002Froot\u002Freference_target_text.srt` with the same parser and pair cues one-to-one by stable subtitle index\u002Forder; fail rather than silently dropping, duplicating, or inventing a cue when the counts or indices do not match.\n\nSynthesize every paired target cue independently and preserve stable cue order. Measure **raw speech duration before adding silence padding, fades, placement, loudness processing, or resampling**. Fit that speech to its assigned window in this order:\n\n1. Use a conservative, bounded speaking-rate adjustment or time-stretch first, then remeasure. Chain bounded tempo factors when needed rather than applying an extreme single factor.\n2. If the speech is shorter than the window, preserve it and append silence to the window boundary.\n3. Trim only as a last resort if the bounded adjustment still cannot fit the window.\n\nRecord the control actually used for each cue as exactly one of `rate_adjust`, `pad_silence`, or `trim`; do not report a planned control if the rendered result used another one. Keep each fitted cue within its assigned window. Apply short fades only after timing edits, and ensure fades do not change the reported placement boundaries. Resampling, cleanup, normalization, and fades must likewise leave the recorded window start\u002Fend and placement calculations unchanged.\n\nCreate one mono WAV per cue in `\u002Foutputs\u002Ftts_segments\u002F`, named in stable index order (`seg_0.wav`, `seg_1.wav`, ...). `seg_0.wav` is the required first-cue artifact, not the complete program timeline. Store the complete assembled timeline separately (for example, `\u002Foutputs\u002Ftts_segments\u002Fassembled_timeline.wav`) with silence outside cue windows, using the integer millisecond offsets from `segments.srt` rather than concatenating clips. The assembled timeline is the audio to normalize and mux; retain the individual cue files for artifact and duration verification.\n\nFor every cue, report the measured raw\u002Ffitted `tts_duration_sec`, `window_start_sec`, `window_end_sec`, actual rendered `placed_start_sec`, `placed_end_sec`, and `drift_sec = placed_end_sec - window_end_sec`. Verify `abs(placed_start_sec - window_start_sec) \u003C= 0.010` seconds and `abs(drift_sec) \u003C= 0.2` seconds before delivery. Render the final assembled audio at 48000 Hz, mono, and measure\u002Fnormalize it using an ITU-R BS.1770-compatible method after timing edits.\n\n\n\n## 5. Multilingual Input, Voice, and Delivery Contract\n\nRead `\u002Froot\u002Ftarget_language.txt` deterministically: use its first non-empty, trimmed line as the target-language specification, normalize it to a language code for reporting, and do not substitute a guessed language. Select a voice that explicitly supports that target language. Prefer an installed offline-capable neural voice; if the preferred engine or voice is unavailable, use an installed offline fallback such as `espeak-ng` with the closest verified target-language voice and record the fallback choice. Never silently synthesize in the source language.\n\nUse the target text from `\u002Froot\u002Freference_target_text.srt` verbatim after parsing; use `\u002Froot\u002Fsource_text.srt` only for source-text pairing and reporting. Preserve cue text and indices, and fail if any required input is missing or pairing is ambiguous. Inspect the generated WAVs and the separately assembled timeline with media metadata tools before writing outputs.\n\nThe complete normalized assembled timeline, not `\u002Foutputs\u002Ftts_segments\u002Fseg_0.wav` unless there is only one cue, must be used to create `\u002Foutputs\u002Fdubbed.mp4` while preserving the original visual stream. The final audio must be 48000 Hz and mono. Write `\u002Foutputs\u002Freport.json` only after validation, using language codes and including the actual measured loudness, original and final durations, audio properties, every paired cue's source\u002Ftarget text, placement boundaries, drift, measured TTS duration, and its actual `duration_control`. Keep the complete timeline artifact separate from the required first-segment artifact and make this distinction explicit in any internal metadata or report.\n",{"name":5,"description":7},"main","cdcb6d230f3b15c3c33f8a2bdd16439f43602cbe","evaluation\u002Fresults\u002FSkillOpt-adapted-gpt52-n3L4R2-v25\u002Ftasks\u002Fmultilingual-video-dubbing\u002Frepaired_skill\u002Ftext-to-speech",1,"https:\u002F\u002Fgithub.com\u002Flinyouxue\u002Fskill-repair-benchmark","https:\u002F\u002Fgithub.com\u002Flinyouxue","2026-11-14 20:28:35",{"source":8,"skillName":5,"commands":32,"prompt":42,"zipAvailable":43},[33,36,39],{"tool":34,"command":35},"npx","npx skills add linyouxue\u002Fskill-repair-benchmark --skill text-to-speech",{"tool":37,"command":38},"bunx","bunx skills add linyouxue\u002Fskill-repair-benchmark --skill text-to-speech",{"tool":40,"command":41},"pnpm","pnpm dlx skills add linyouxue\u002Fskill-repair-benchmark --skill text-to-speech","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Flinyouxue\u002Fskill-repair-benchmark\u002Fblob\u002Fmain\u002Fevaluation\u002Fresults\u002FSkillOpt-adapted-gpt52-n3L4R2-v25\u002Ftasks\u002Fmultilingual-video-dubbing\u002Frepaired_skill\u002Ftext-to-speech\u002FSKILL.md",true,[],[46,53,60,67,72,77],{"skillNo":47,"name":48,"nameZh":6,"description":49,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":50,"categories":51,"keywords":52,"publishedAt":19},"01M7E1HGVRY73NBJ022CJG13X6","video-frame-extraction","Extract frames from video files and save them as images using OpenCV","https:\u002F\u002Fgithub.com\u002Flinyouxue\u002Fskill-repair-benchmark\u002Fblob\u002Fmain\u002Fjpg-ocr-stat\u002Fresult\u002Fskilltta-claude4.7\u002Fgroup_03\u002Frun_03\u002Finputs\u002Fskills\u002Fvideo-frame-extraction\u002FSKILL.md",[],[],{"skillNo":54,"name":55,"nameZh":6,"description":56,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":57,"categories":58,"keywords":59,"publishedAt":19},"01M7E1HAG5R7CJ9C8NYJS771MB","openai-vision","Analyze images and multi-frame sequences using OpenAI GPT vision models","https:\u002F\u002Fgithub.com\u002Flinyouxue\u002Fskill-repair-benchmark\u002Fblob\u002Fmain\u002Fjpg-ocr-stat\u002Fresult\u002Fskilltta-claude4.7\u002Fgroup_03\u002Frun_03\u002Finputs\u002Fskills\u002Fopenai-vision\u002FSKILL.md",[],[],{"skillNo":61,"name":62,"nameZh":6,"description":63,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":64,"categories":65,"keywords":66,"publishedAt":19},"01M7E1H7DADKB6QBWKC2KQ63SZ","image-ocr","Extract text content from images using Tesseract OCR via Python, including structured field extraction from receipts and forms","https:\u002F\u002Fgithub.com\u002Flinyouxue\u002Fskill-repair-benchmark\u002Fblob\u002Fmain\u002Fjpg-ocr-stat\u002Fresult\u002Fskilltta-claude4.7\u002Fgroup_03\u002Frun_03\u002Finputs\u002Fskills\u002Fimage-ocr\u002FSKILL.md",[],[],{"skillNo":68,"name":48,"nameZh":6,"description":49,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":69,"categories":70,"keywords":71,"publishedAt":19},"01M7E1H0WBNT43TENGQY7AVZHD","https:\u002F\u002Fgithub.com\u002Flinyouxue\u002Fskill-repair-benchmark\u002Fblob\u002Fmain\u002Fjpg-ocr-stat\u002Fresult\u002Fskilltta-claude4.7\u002Fgroup_03\u002Frun_02\u002Finputs\u002Fskills\u002Fvideo-frame-extraction\u002FSKILL.md",[],[],{"skillNo":73,"name":55,"nameZh":6,"description":56,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":74,"categories":75,"keywords":76,"publishedAt":19},"01M7E1GTC7D4N4Q2EF03875J9E","https:\u002F\u002Fgithub.com\u002Flinyouxue\u002Fskill-repair-benchmark\u002Fblob\u002Fmain\u002Fjpg-ocr-stat\u002Fresult\u002Fskilltta-claude4.7\u002Fgroup_03\u002Frun_02\u002Finputs\u002Fskills\u002Fopenai-vision\u002FSKILL.md",[],[],{"skillNo":78,"name":62,"nameZh":6,"description":63,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":79,"categories":80,"keywords":81,"publishedAt":19},"01M7E1GQ92CWFMGH53KSSE4WER","https:\u002F\u002Fgithub.com\u002Flinyouxue\u002Fskill-repair-benchmark\u002Fblob\u002Fmain\u002Fjpg-ocr-stat\u002Fresult\u002Fskilltta-claude4.7\u002Fgroup_03\u002Frun_02\u002Finputs\u002Fskills\u002Fimage-ocr\u002FSKILL.md",[],[]]