[{"data":1,"prerenderedAt":1633},["ShallowReactive",2],{"guides-en":3},[4,275,487,680,927,1180,1388],{"id":5,"title":6,"author":7,"body":8,"date":260,"description":261,"difficulty":262,"extension":263,"image":264,"lastModified":260,"meta":265,"navigation":266,"outcome":267,"path":268,"readingTime":269,"robots":264,"scenario":270,"seo":271,"stem":272,"tools":273,"__hash__":274},"guides\u002F8.guides\u002F1.archive-a-course-over-hls.md","How to Keep a Course That Only Streams","FlowPick Team",{"type":9,"value":10,"toc":248},"minimark",[11,15,18,23,26,29,37,40,44,55,72,79,84,88,100,106,111,114,127,131,139,145,150,153,156,160,163,166,172,178,181,185,188,196,199,203,213,224,230,234,241],[12,13,14],"p",{},"Course platforms rarely hand you a file. They stream segments, and the segments are listed in a playlist you never see in the address bar. This guide walks through turning one of those playlists into a folder of usable MP4s.",[12,16,17],{},"It is written from the problems we hit while doing this, not from the documentation of the format. Where something did not work for us, that is in the text.",[19,20,22],"h2",{"id":21},"first-does-this-apply-to-your-course","First, does this apply to your course?",[12,24,25],{},"Two conditions have to hold.",[12,27,28],{},"You can already play the video in your own browser, and the platform does not encrypt the segments with DRM. If the video is behind Widevine or FairPlay, the segments are ciphertext and the decryption keys never leave the player's protected memory. FlowPick does not attempt to break that, and no browser-side tool can.",[12,30,31,32,36],{},"The quick check: open DevTools, go to the Network panel, filter for ",[33,34,35],"code",{},"m3u8",", and start the video. If a playlist request appears and the segment responses are readable media, the stream is unencrypted and this workflow applies. If you only see segment requests with no playlist, or the responses look like noise, stop here — this guide will not help.",[12,38,39],{},"Also worth stating plainly: this workflow is for material you are allowed to keep, such as your own recordings or courses whose terms permit offline copies. Whether a specific platform allows it is in that platform's terms, not in this guide.",[19,41,43],{"id":42},"why-the-playlist-is-the-part-that-matters","Why the playlist is the part that matters",[12,45,46,47,54],{},"HLS is specified in ",[48,49,53],"a",{"href":50,"rel":51},"https:\u002F\u002Fwww.rfc-editor.org\u002Frfc\u002Frfc8216",[52],"nofollow","RFC 8216",", and the structure it defines is the reason this workflow has two steps where you would expect one.",[12,56,57,58,62,63,67,68,71],{},"A ",[59,60,61],"strong",{},"master playlist"," does not contain media. It lists ",[64,65,66],"em",{},"variants",", each with its own resolution and its own ",[59,69,70],{},"media playlist",", which in turn lists the actual segments. Downloading \"the video\" therefore means choosing a variant first, whether you do it deliberately or let a tool do it for you.",[12,73,74],{},[75,76],"img",{"alt":77,"src":78},"HLS playlist structure: a master playlist points to three variants, each with its own media playlist of segments","\u002Fimg\u002Fguides\u002Fhls-playlist-structure.svg",[12,80,81],{},[64,82,83],{},"One master playlist, three variants, one segment list per variant. The highlighted variant is the one this guide downloads.",[19,85,87],{"id":86},"step-1-read-the-playlist-before-you-download-anything","Step 1: Read the playlist before you download anything",[12,89,90,91,94,95,99],{},"Copy the ",[33,92,93],{},".m3u8"," URL from the Network panel and paste it into the ",[48,96,98],{"href":97},"\u002Ftools\u002Fm3u8-playlist-analyzer","playlist analyser",". The download step requires the FlowPick browser extension; install it in a supported browser before continuing.",[12,101,102],{},[75,103],{"alt":104,"src":105},"Five video variants in the Mux public HLS example, showing resolution, bandwidth and codecs","\u002Fimg\u002Fguides\u002Fplaylist-analyzer-variants.jpg",[12,107,108],{},[64,109,110],{},"Public Mux test stream, not a course recording. The master lists five variants; open a variant media playlist separately to inspect its segments.",[12,112,113],{},"The analyser lists the variants, declared resolutions and codecs in a master playlist, along with any separate audio entries. Open the chosen variant’s media playlist separately to inspect segment counts and listed duration. A master playlist itself has no media segments. That matters more than it sounds:",[115,116,117,121,124],"ul",{},[118,119,120],"li",{},"A course recorded in 2020 is often stored at 720p while the master also advertises a 1080p variant that was upscaled from it. Downloading the biggest variant gets you a larger file with no extra detail.",[118,122,123],{},"Separate audio tracks are common on newer platforms. If you hand the master playlist straight to a downloader, audio handling becomes something you find out about after the fact.",[118,125,126],{},"Segment count tells you roughly how long the download will take and whether the browser will hold up. A two-hour lecture is several hundred segments.",[19,128,130],{"id":129},"step-2-give-the-downloader-one-variant-not-the-master","Step 2: Give the downloader one variant, not the master",[12,132,133,134,138],{},"Take the URL of the variant you picked and paste it into the ",[48,135,137],{"href":136},"\u002Fm3u8-downloader","M3U8 downloader",".",[12,140,141],{},[75,142],{"alt":143,"src":144},"HLS downloader fetching media segments with progress, segment count and speed","\u002Fimg\u002Fguides\u002Fm3u8-downloader-progress.jpg",[12,146,147],{},[64,148,149],{},"Real download of the public Mux 720p test variant in Chrome with FlowPick installed. This is a progress snapshot, not a completed course archive.",[12,151,152],{},"Passing a single variant instead of the master is deliberately more work, because it removes a decision from the tool: the segment list is fixed, so a failure points at a specific segment instead of at variant selection.",[12,154,155],{},"If the playlist needs a referer or cookie to serve segments, the extension route usually works better than the standalone page, since the extension attaches the request context from the tab that is already playing the video.",[19,157,159],{"id":158},"step-3-merge-and-watch-for-the-two-common-stops","Step 3: Merge and watch for the two common stops",[12,161,162],{},"The downloader fetches the segments and assembles them into a single MP4 in the browser. Nothing is uploaded.",[12,164,165],{},"Two things interrupt this step in practice:",[12,167,168,171],{},[59,169,170],{},"Signed segment URLs expiring mid-download."," Many CDNs sign each segment URL with a short expiry. A 40-minute download can outlive them, and the tail of the file 403s. Re-copying the playlist gets a fresh signature; a shorter variant or a smaller range gets you under the expiry window.",[12,173,174,177],{},[59,175,176],{},"Browser storage limits."," A long course is gigabytes. Chrome and Edge write through the file system, so the limit is your disk. Firefox stages into browser storage and asks you to save the result afterwards, which needs free space in both places at once. Close other tabs before starting a multi-hour job.",[12,179,180],{},"Keep the tab open until it finishes. Backgrounding is fine; closing is not.",[19,182,184],{"id":183},"step-4-trim-what-you-will-never-rewatch","Step 4: Trim what you will never rewatch",[12,186,187],{},"Recorded lectures tend to carry 20–40 seconds of title card at the front and a slide of \"thank you for watching\" at the end. Multiply that by a semester and it is worth removing.",[12,189,190,191,195],{},"The ",[48,192,194],{"href":193},"\u002Ftools\u002Fvideo-trimmer","video trimmer"," cuts by timestamp. Work from the merged file rather than from segments — one operation instead of several, and you keep the original until you are satisfied with the cut.",[12,197,198],{},"Check the result for two things before deleting the original: that the audio does not drift out of sync near the cut, and that the first second after the cut is not a frozen frame. Both are consequences of cutting near a keyframe rather than at an arbitrary point.",[19,200,202],{"id":201},"where-people-get-stuck","Where people get stuck",[12,204,205,208,209,212],{},[59,206,207],{},"The analyser shows one variant but downloads fail."," Check whether the playlist requires a specific ",[33,210,211],{},"Referer"," header. Some course platforms reject requests that do not look like they came from the player page.",[12,214,215,218,219,223],{},[59,216,217],{},"The merged file plays but has no sound."," The audio was a separate track that did not get merged. Go back to the analyser, look for a separate audio entry in the master playlist, and check whether the downloader has an option for it. If it does not, download the audio variant through the ",[48,220,222],{"href":221},"\u002Faudio-downloader","audio downloader"," and keep the two files together.",[12,225,226,229],{},[59,227,228],{},"Every segment 404s after the first minute."," The playlist is stale. Re-open the lecture in the player, re-copy the current playlist URL, and start again.",[19,231,233],{"id":232},"what-you-end-up-with","What you end up with",[12,235,236,237,240],{},"One MP4 per lecture at a resolution you chose, with the padding trimmed. Name the files so the ordering survives: ",[33,238,239],{},"course-03-hashing.mp4"," sorts better than a timestamp, and you will be glad of it two years from now.",[12,242,190,243,247],{},[48,244,246],{"href":245},"\u002Fdocs\u002Ffeatures\u002Fformat-conversion","format conversion"," page covers what the merge step does not change — container, codecs, and why a 720p source stays 720p.",{"title":249,"searchDepth":250,"depth":250,"links":251},"",2,[252,253,254,255,256,257,258,259],{"id":21,"depth":250,"text":22},{"id":42,"depth":250,"text":43},{"id":86,"depth":250,"text":87},{"id":129,"depth":250,"text":130},{"id":158,"depth":250,"text":159},{"id":183,"depth":250,"text":184},{"id":201,"depth":250,"text":202},{"id":232,"depth":250,"text":233},"2026-10-29","Course platforms stream segments from a playlist you never see. Read the playlist first, pick one variant instead of the master, merge, then trim. A worked example with the failure points named.","intermediate","md",null,{},true,"One MP4 per lecture, at a resolution you chose, with the intro and outro removed.","\u002Fguides\u002Farchive-a-course-over-hls",8,"archive",{"title":6,"description":261},"8.guides\u002F1.archive-a-course-over-hls",[97,136,193],"j8JMIaTKlNfBT4QwBoZgnRVaZqLPWFOtfU0KDZ9zmt4",{"id":276,"title":277,"author":7,"body":278,"date":260,"description":476,"difficulty":477,"extension":263,"image":264,"lastModified":260,"meta":478,"navigation":266,"outcome":479,"path":480,"readingTime":481,"robots":264,"scenario":482,"seo":483,"stem":484,"tools":485,"__hash__":486},"guides\u002F8.guides\u002F2.turn-a-meeting-recording-into-notes.md","How to Turn a Meeting into Searchable Text",{"type":9,"value":279,"toc":467},[280,283,287,290,293,296,300,306,312,317,320,326,331,334,338,341,349,352,358,363,367,375,378,384,390,396,401,404,408,411,417,422,425,429,435,441,447,453,455,458],[12,281,282],{},"Ten minutes after a meeting ends, nobody remembers what was decided. The recording exists, but scrubbing a 70-minute video to find one number is worse than having no recording. This guide turns the file into something searchable.",[19,284,286],{"id":285},"what-this-workflow-does-and-does-not-do","What this workflow does and does not do",[12,288,289],{},"It produces a transcript: timestamped text, editable, exportable as TXT, SRT or VTT. That is the raw material for minutes.",[12,291,292],{},"It does not summarise. There is no \"summarise this meeting\" button, and the transcription tools do not translate or identify speakers. If you need a summary, you write it from the transcript — which is usually faster than writing it from the video, because you can skim.",[12,294,295],{},"One constraint worth knowing before you start: recognition runs on your own machine, so duration is cost. In our runs, a 70-minute meeting on a laptop CPU took long enough that trimming was not optional.",[19,297,299],{"id":298},"step-1-trim-before-you-transcribe","Step 1: Trim before you transcribe",[12,301,302,303,305],{},"Open the recording in the ",[48,304,194],{"href":193}," and cut two things: the stretch at the front where people join and check their audio, and the tail after the last decision.",[12,307,308],{},[75,309],{"alt":310,"src":311},"A 70-minute recording split into 9 minutes of joining, 45 minutes of agenda and 15 minutes of closing chat; only the agenda is kept","\u002Fimg\u002Fguides\u002Ftrim-before-transcribe.svg",[12,313,314],{},[64,315,316],{},"Recognition cost scales with duration. Cutting the first nine minutes removes 13% of the work before it starts.",[12,318,319],{},"This is the highest-leverage step, and the reason is arithmetic rather than tidiness: a meeting that starts nine minutes late becomes 13% shorter after the cut, and the parts you removed contain nothing you need.",[12,321,322],{},[75,323],{"alt":324,"src":325},"Video trimmer with a selection from one to three seconds and a two-second output duration","\u002Fimg\u002Fguides\u002Fvideo-trimmer-timeline.jpg",[12,327,328],{},[64,329,330],{},"Interface example using the four-second FlowPick test video. The highlighted range keeps 0:01–0:03; apply the same controls to your meeting.",[12,332,333],{},"Keep the original. Trimming produces a new file, and you will want to go back to the untouched version if a decision turns out to be ambiguous.",[19,335,337],{"id":336},"step-2-convert-webm-to-mp4-only-if-transcription-fails","Step 2: Convert WebM to MP4 only if transcription fails",[12,339,340],{},"Browser screen recorders, and several meeting clients, write WebM with Opus audio. That is a perfectly good format, and the transcriber often handles it — but \"often\" is doing real work in that sentence, because container and audio decoding support differ between Chrome, Edge and Firefox, and the audio track is what matters here, not the picture.",[12,342,343,344,348],{},"If the transcriber reports that it cannot decode the audio, or silently finds no speech, convert first with the ",[48,345,347],{"href":346},"\u002Ftools\u002Fwebm-to-mp4","WebM to MP4 converter"," and try the MP4.",[12,350,351],{},"Skip this step if transcription already works. Converting a file that was going to be fine costs a few minutes for nothing.",[12,353,354],{},[75,355],{"alt":356,"src":357},"WebM conversion result showing an 88.9 KB VP8 Opus source and a 122.3 KB MP4 output","\u002Fimg\u002Fguides\u002Fwebm-to-mp4-result.jpg",[12,359,360],{},[64,361,362],{},"Generated four-second test video: 88.9 KB WebM → 122.3 KB MP4. This tool re-encodes to H.264\u002FAAC; output size depends on the source and settings.",[19,364,366],{"id":365},"step-3-transcribe-choosing-the-model-deliberately","Step 3: Transcribe, choosing the model deliberately",[12,368,369,370,374],{},"Open ",[48,371,373],{"href":372},"\u002Ftools\u002Fvideo-to-text","Video to Text",", select the trimmed MP4, set the spoken language, and start.",[12,376,377],{},"Two settings decide how long this takes:",[12,379,380,383],{},[59,381,382],{},"Model size."," The lightweight model downloads 40–150 MB of weights and is the right choice for a meeting where you only need to find decisions. The standard model downloads 80–250 MB and recognises more, especially names and technical terms, at the cost of time and memory.",[12,385,386,389],{},[59,387,388],{},"Processing mode."," GPU acceleration helps if the browser exposes it. Auto mode falls back to CPU when it cannot confirm a usable GPU, which is common on laptops with integrated graphics.",[12,391,392],{},[75,393],{"alt":394,"src":395},"Video transcription settings with spoken language, Whisper Tiny and Base model choices, and CPU processing","\u002Fimg\u002Fguides\u002Fvideo-to-text-settings.jpg",[12,397,398],{},[64,399,400],{},"Transcription settings for the generated English speech sample. Model choice and processing mode affect download size and processing time.",[12,402,403],{},"Language is the setting people get wrong most often. Auto-detection reads the first stretch with speech and applies that for the rest; a meeting that opens with two people chatting in another language will be misdetected. If you know the language, set it.",[19,405,407],{"id":406},"step-4-correct-the-parts-that-matter","Step 4: Correct the parts that matter",[12,409,410],{},"Click a timestamp to play the video from that point, then fix the text. You do not need to correct everything. Names, numbers, dates and action items are worth the effort; filler is not.",[12,412,413],{},[75,414],{"alt":415,"src":416},"Video transcript with an editable segment, a zero-to-five-second timestamp and TXT SRT VTT export buttons","\u002Fimg\u002Fguides\u002Ftranscript-editing.jpg",[12,418,419],{},[64,420,421],{},"Actual Whisper Tiny CPU output from the generated speech sample, before correction. It misheard “two short sentences” as “new short sentences”; review the audio before quoting.",[12,423,424],{},"Timestamps are approximate. If the transcript is going to become subtitles, plan on adjusting the timing in your editor rather than trusting it exactly.",[19,426,428],{"id":427},"where-this-goes-wrong","Where this goes wrong",[12,430,431,434],{},[59,432,433],{},"Music at the start confuses detection."," Overlapping speakers and background music both hurt recognition. Trimming the intro in step 1 solves most of it.",[12,436,437,440],{},[59,438,439],{},"The transcript is a wall of text with no line breaks."," The output is segmented by recognition windows, not by speaker or sentence. Expect to restructure it when writing minutes.",[12,442,443,446],{},[59,444,445],{},"A long recording stops partway."," Completed sections are kept. You can export what exists, or split the source file and transcribe the halves separately — which is also the workaround for a browser that runs out of memory.",[12,448,449,452],{},[59,450,451],{},"Someone spoke in a language the model handles poorly."," This version supports the languages listed in the tool, and quality varies by language. Check the output before quoting anything.",[19,454,233],{"id":232},[12,456,457],{},"Three files from one session: the trimmed MP4 for anyone who wants to watch, an editable transcript for grepping, and an SRT if you need captions.",[12,459,460,461,463,464,138],{},"For the subtitle-specific path — burning them in, checking timing on a timeline — see ",[48,462,246],{"href":245}," and the ",[48,465,466],{"href":372},"SRT export notes",{"title":249,"searchDepth":250,"depth":250,"links":468},[469,470,471,472,473,474,475],{"id":285,"depth":250,"text":286},{"id":298,"depth":250,"text":299},{"id":336,"depth":250,"text":337},{"id":365,"depth":250,"text":366},{"id":406,"depth":250,"text":407},{"id":427,"depth":250,"text":428},{"id":232,"depth":250,"text":233},"Screen recordings arrive as WebM with the waiting at the front. Trim first, convert only if the audio will not decode, then transcribe. Includes what recognition does badly and how long it takes.","basic",{},"A trimmed MP4 for sharing, plus an editable transcript and SRT file from the same run.","\u002Fguides\u002Fturn-a-meeting-recording-into-notes",7,"transcribe",{"title":277,"description":476},"8.guides\u002F2.turn-a-meeting-recording-into-notes",[193,346,372],"WCxWAx5vHs1M4oB2YcAUKKdWrdKdLbkTlRtGxgzeTfQ",{"id":488,"title":489,"author":7,"body":490,"date":260,"description":670,"difficulty":477,"extension":263,"image":264,"lastModified":260,"meta":671,"navigation":266,"outcome":672,"path":673,"readingTime":674,"robots":264,"scenario":675,"seo":676,"stem":677,"tools":678,"__hash__":679},"guides\u002F8.guides\u002F3.check-a-file-that-will-not-play.md","Why a File That Won't Play Is Usually Two Problems",{"type":9,"value":491,"toc":662},[492,495,499,502,508,514,517,523,528,531,534,538,546,552,557,560,586,589,593,601,608,613,618,622,625,628,631,634,638,649,651,654],[12,493,494],{},"Someone sends you a video. It will not open. The instinct is to convert it to MP4. Half the time that fixes it, and half the time you spend twenty minutes converting a file whose real problem was never the container.",[19,496,498],{"id":497},"two-problems-that-look-identical","Two problems that look identical",[12,500,501],{},"A media file has two layers that matter here.",[12,503,190,504,507],{},[59,505,506],{},"container"," is the box: MP4, WebM, MKV, TS, MOV. It says where the streams live, not what they are. Each format is a published specification — MP4 is standardised as ISO\u002FIEC 14496-12, Matroska and WebM have their own.",[12,509,190,510,513],{},[59,511,512],{},"codec"," is what is inside: H.264, H.265, VP9, AV1 for video; AAC, Opus, AC-3 for audio.",[12,515,516],{},"A player that cannot open a file is usually reporting a container problem, because it cannot even find the streams to try decoding them. But a player that opens the file and then shows a black frame or silent audio has a codec problem, and no amount of container swapping will help.",[12,518,519],{},[75,520],{"alt":521,"src":522},"Comparing containers and codecs: MKV and MP4 holding the same H.264 and AAC streams can be remuxed, while an MP4 holding H.265 needs a re-encode","\u002Fimg\u002Fguides\u002Fcontainer-vs-codec.svg",[12,524,525],{},[64,526,527],{},"Changing the box is cheap and lossless. Changing what is inside is neither.",[12,529,530],{},"Converting the container is called remuxing. It copies the encoded streams into a new box. It is fast, it needs no quality loss, and it is the right answer when the box is the problem.",[12,532,533],{},"Re-encoding decodes every frame and compresses it again. It is slow, it loses a little quality, and it is the only answer when the codec is the problem.",[19,535,537],{"id":536},"step-1-read-the-file-instead-of-guessing","Step 1: Read the file instead of guessing",[12,539,540,541,545],{},"Drop the file into ",[48,542,544],{"href":543},"\u002Ftools\u002Fvideo-info","Video Info",". It reports the container, the video and audio codecs, resolution, duration and track layout.",[12,547,548],{},[75,549],{"alt":550,"src":551},"Media report showing MP4 container, four-second duration, AVC video and AAC audio tracks","\u002Fimg\u002Fguides\u002Fvideo-info-output.jpg",[12,553,554],{},[64,555,556],{},"Generated four-second MP4 with AVC\u002FH.264 video and AAC audio. Unavailable bitrate fields are shown honestly rather than estimated.",[12,558,559],{},"What you are looking for:",[115,561,562,568,574,580],{},[118,563,564,567],{},[59,565,566],{},"Container is MKV or TS, codecs are H.264 + AAC."," Your player simply does not handle the box. Remux to MP4 and you are done in a minute or two.",[118,569,570,573],{},[59,571,572],{},"Container is MP4, video codec is H.265."," The box is fine; the player lacks an H.265 decoder. Remuxing will not help. You need an actual re-encode to a codec your player supports. The WebM converter targets WebM input; for an H.265 MP4, use an encoder that supports that source format.",[118,575,576,579],{},[59,577,578],{},"Audio codec is AC-3 or DTS."," Common in files ripped from discs. Some browsers and phones will not play them even in an MP4 container. The fix is re-encoding the audio, not the video.",[118,581,582,585],{},[59,583,584],{},"No audio track at all."," Not a format problem. The recording was silent.",[12,587,588],{},"That last case catches people regularly: a \"broken video\" that simply has no sound because the capture tool did not record any.",[19,590,592],{"id":591},"step-2-choose-between-remuxing-and-re-encoding","Step 2: Choose between remuxing and re-encoding",[12,594,595,596,600],{},"For a TS file containing compatible H.264 video and AAC audio, use ",[48,597,599],{"href":598},"\u002Ftools\u002Fts-to-mp4","TS to MP4",". It copies the encoded streams into MP4 without re-encoding. The file may become smaller because the new container has different overhead.",[12,602,603,604,607],{},"For a browser recording in WebM, use ",[48,605,606],{"href":346},"WebM to MP4"," when you need H.264\u002FAAC compatibility. This tool decodes and re-encodes the source; it does not simply swap the container. Conversion takes processing time, may lose some quality, and can increase or decrease file size.",[12,609,610],{},[75,611],{"alt":612,"src":357},"WebM source and MP4 result with different file sizes after re-encoding",[12,614,615],{},[64,616,617],{},"WebM conversion is a re-encode, not the TS remux described above. This generated sample grows from 88.9 KB to 122.3 KB.",[19,619,621],{"id":620},"step-3-know-when-to-stop","Step 3: Know when to stop",[12,623,624],{},"There is a category of file this workflow cannot rescue, and recognising it saves an afternoon.",[12,626,627],{},"Encrypted streams are one. If the source was DRM-protected, the segments are ciphertext and there is no container operation that produces playable video.",[12,629,630],{},"Codecs your target device lacks are the other. An H.265 file will not become playable on an H.264-only device by changing its extension or its container. It needs re-encoding, which is a different and slower job.",[12,632,633],{},"Stating it plainly is more useful than a workaround that does not exist: if the file needs a codec the player does not have, the fix is on the player side or in a real encoder.",[19,635,637],{"id":636},"a-note-on-extensions","A note on extensions",[12,639,640,641,644,645,648],{},"Renaming ",[33,642,643],{},"video.webm"," to ",[33,646,647],{},"video.mp4"," does not convert anything. The container is described inside the file, not by its name. Players that trust the extension will fail in confusing ways, and some will refuse the file outright. This is worth mentioning because it is the most common thing people try first.",[19,650,233],{"id":232},[12,652,653],{},"Either a file that opens everywhere because the container was the problem, or a clear answer that the codec is the problem and a re-encode is required.",[12,655,656,657,661],{},"For the layers involved — tracks, segments, and why a remux preserves them — the ",[48,658,660],{"href":659},"\u002Fdocs\u002Fadvanced\u002Fdownload-engine","download engine"," page goes deeper.",{"title":249,"searchDepth":250,"depth":250,"links":663},[664,665,666,667,668,669],{"id":497,"depth":250,"text":498},{"id":536,"depth":250,"text":537},{"id":591,"depth":250,"text":592},{"id":620,"depth":250,"text":621},{"id":636,"depth":250,"text":637},{"id":232,"depth":250,"text":233},"A container the player does not know and a codec it cannot decode need different fixes. How to read a file, tell the two apart, and when the honest answer is that a re-encode is the only way.",{},"A correct diagnosis, followed by a compatible TS remux, a WebM re-encode, or a clear next step for unsupported codecs.","\u002Fguides\u002Fcheck-a-file-that-will-not-play",6,"convert",{"title":489,"description":670},"8.guides\u002F3.check-a-file-that-will-not-play",[543,346],"uyL5WszuRsBjjgULpELSPR1cP2AGvpmGeaOrW9CJnWA",{"id":681,"title":682,"author":7,"body":683,"date":260,"description":919,"difficulty":262,"extension":263,"image":264,"lastModified":260,"meta":920,"navigation":266,"outcome":921,"path":922,"readingTime":481,"robots":264,"scenario":270,"seo":923,"stem":924,"tools":925,"__hash__":926},"guides\u002F8.guides\u002F4.save-a-live-replay.md","How to Save a Live Replay Before It Expires",{"type":9,"value":684,"toc":910},[685,688,691,695,698,704,710,716,722,727,731,740,765,771,776,780,786,789,795,801,804,808,811,817,838,844,849,856,860,863,868,876,878,884,890,896,905,907],[12,686,687],{},"A live replay is the least durable form of video on the internet. Platforms keep it for a week, a month, or until someone changes a setting, and then it is gone. If you need it later — for a report, a highlight, a compliance record — \"later\" is usually after it has been deleted.",[12,689,690],{},"This guide is about capturing it while it exists, and about the fact that a replay is a harder source than an ordinary video.",[19,692,694],{"id":693},"why-a-replay-is-not-just-a-video","Why a replay is not just a video",[12,696,697],{},"Three things differ from a normal video-on-demand file, and each one changes what you do.",[12,699,700,703],{},[59,701,702],{},"There is a deadline."," Replays expire on a platform-defined schedule. You cannot download it again next month, so the copy you make is the copy you keep. That changes the quality bar: getting a usable file once matters more than getting the optimal file.",[12,705,706,709],{},[59,707,708],{},"The stream is more segmented."," Live delivery cuts segments far shorter than a VOD file does, often two to six seconds instead of ten. A three-hour replay can be several thousand segments. Segment count is the main reason a capture fails partway.",[12,711,712,715],{},[59,713,714],{},"The bitrate moves."," Live encoding adapts to what the encoder can push, so quality varies through the recording. A replay of a stream that struggled at the twenty-minute mark will look worse there, and no post-processing fixes it.",[12,717,718],{},[75,719],{"alt":720,"src":721},"Timeline of a stream: live, then a replay window set by the platform, then deleted","\u002Fimg\u002Fguides\u002Freplay-capture-window.svg",[12,723,724],{},[64,725,726],{},"The replay window is the only time this workflow is possible, and its length is not yours to decide.",[19,728,730],{"id":729},"step-1-check-what-the-page-actually-gives-you","Step 1: Check what the page actually gives you",[12,732,733,734,736,737,138],{},"Before recording, spend thirty seconds looking at the Network panel. Filter for ",[33,735,35],{}," and ",[33,738,739],{},"mpd",[115,741,742,748,759],{},[118,743,744,747],{},[59,745,746],{},"A playlist appears."," The stream is unencrypted and listed. You have a choice between capturing the media data and recording the screen; capturing is better because it copies the encoded streams rather than re-encoding a picture of them.",[118,749,750,753,754,758],{},[59,751,752],{},"No playlist, but the video plays."," The page may be using Media Source Extensions to feed segments the player fetched itself. The ",[48,755,757],{"href":756},"\u002Frecording","recording tool"," can take the media data in that case, which still avoids re-encoding.",[118,760,761,764],{},[59,762,763],{},"Segment requests but no playlist, and responses look like noise."," DRM. Stop here. Capturing ciphertext produces a file that will not play, and this guide is not about that.",[12,766,767],{},[75,768],{"alt":769,"src":770},"Browser Network panel filtered for m3u8, showing a master playlist and a variant playlist","\u002Fscreenshots\u002Fnetwork-filter-m3u8.png",[12,772,773],{},[64,774,775],{},"Existing project screenshot of the public Mux test stream. Filtering for m3u8 shows playlists; use ts or m4s to inspect segment requests separately.",[19,777,779],{"id":778},"step-2-capture-and-decide-about-the-tail","Step 2: Capture, and decide about the tail",[12,781,782,783,785],{},"Open the replay in the player, click the FlowPick extension icon, and choose its recording control. Select MSE buffer mode when available; the extension opens the ",[48,784,757],{"href":756}," with the active session. Opening that URL directly shows instructions rather than starting a recording.",[12,787,788],{},"Two practical notes:",[12,790,791,794],{},[59,792,793],{},"Do not record faster than real time if the source is a live stream."," A replay that is still being written has no future segments to fetch. If the replay is complete, capturing the media data can run faster than playback; if it is still live, you are bounded by the clock.",[12,796,797,800],{},[59,798,799],{},"Leave the tab alone."," Capture runs in the page. Switching tabs is fine in most browsers, but closing, sleeping the machine, or letting the browser discard the tab ends the recording.",[12,802,803],{},"If the replay is four hours long and you only need one match, consider a different approach: record only during the part you need. It is less work than capturing everything and cutting later, and it halves the chance of hitting a segment failure three hours in.",[19,805,807],{"id":806},"step-3-fix-the-container-not-the-picture","Step 3: Fix the container, not the picture",[12,809,810],{},"What comes out of a capture depends on what went in. Live streams are commonly delivered as fragmented MP4 or as MPEG-TS.",[12,812,813,814,816],{},"If the file plays everywhere, skip this step. If it does not, read it first with ",[48,815,544],{"href":543},", then decide:",[115,818,819,828],{},[118,820,821,824,825,827],{},[59,822,823],{},"TS container, H.264 + AAC codecs."," Convert with ",[48,826,599],{"href":598},". This copies the streams into a better-supported wrapper; it does not re-encode, so it is quick and loses nothing.",[118,829,830,833,834,837],{},[59,831,832],{},"Already MP4 and still unplayable."," Inspect the codec and file integrity before choosing a fix. A codec your player cannot decode needs a re-encode; the TS remux cannot change it. Check the ",[48,835,836],{"href":673},"container and codec guide"," before spending time on it.",[12,839,840],{},[75,841],{"alt":842,"src":843},"TS to MP4 result showing 35.8 KB input and 31.4 KB output","\u002Fimg\u002Fguides\u002Fts-to-mp4-sizes.jpg",[12,845,846],{},[64,847,848],{},"Actual remux of the generated TS sample: 35.8 KB input → 31.4 KB MP4. Container overhead changes file size; verify duration and tracks in Video Info.",[12,850,851,852,855],{},"A note on sizes that surprises people: a TS to MP4 remux often produces a ",[64,853,854],{},"smaller"," file even though nothing was re-encoded. TS carries packet overhead and padding that the MP4 container does not need. A 15% drop is normal and does not mean quality was lost.",[19,857,859],{"id":858},"step-4-cut-it-down-before-you-archive-it","Step 4: Cut it down before you archive it",[12,861,862],{},"A three-hour replay of a match, a conference, or a live stream is usually watched once. Keeping all of it costs disk and makes it hard to find the part that mattered.",[12,864,190,865,867],{},[48,866,194],{"href":193}," cuts by timestamp. Two habits worth having:",[115,869,870,873],{},[118,871,872],{},"Keep the original until the cut is verified. Trimmed output is a new file; the capture is the only copy of something that may no longer exist online.",[118,874,875],{},"Cut on chapter boundaries where the platform provided them, not mid-sentence. It makes the result easier to name and easier to find later.",[19,877,428],{"id":427},[12,879,880,883],{},[59,881,882],{},"The capture stops around the segment limit."," Several thousand segments is a lot of memory. Split the job: record the first half, then the second. Two files that combine cleanly beat one file that failed at 80%.",[12,885,886,889],{},[59,887,888],{},"Audio and video drift apart in a long capture."," This is a live-encoding artefact more than a tool failure, and it gets worse with duration. Shorter captures reduce it; there is no reliable post-fix in the browser.",[12,891,892,895],{},[59,893,894],{},"The replay expires mid-capture."," Nothing recovers the missing tail. This is the one failure that argues for capturing early rather than waiting for a convenient moment.",[12,897,898,901,902,904],{},[59,899,900],{},"The result plays but has no audio."," The audio was a separate track that did not get picked up. Re-capture with the audio track selected, or capture the audio separately with the ",[48,903,222],{"href":221}," if the source has a separate audio playlist.",[19,906,233],{"id":232},[12,908,909],{},"A local file that no longer depends on the platform's retention policy, in a container that opens in a player and an editor. Trim it down, name it with the date and the event, and check that it plays from a cold start — some captures play from cache but fail on a fresh load, and you want to find that out now rather than in six months.",{"title":249,"searchDepth":250,"depth":250,"links":911},[912,913,914,915,916,917,918],{"id":693,"depth":250,"text":694},{"id":729,"depth":250,"text":730},{"id":778,"depth":250,"text":779},{"id":806,"depth":250,"text":807},{"id":858,"depth":250,"text":859},{"id":427,"depth":250,"text":428},{"id":232,"depth":250,"text":233},"Platforms delete replays on a schedule, and a replay is a worse source than a video on demand: variable bitrate, more segments, and a capture window you do not control. How to get a file while it still exists.",{},"A local copy of the replay, in a container your player and editor both accept.","\u002Fguides\u002Fsave-a-live-replay",{"title":682,"description":919},"8.guides\u002F4.save-a-live-replay",[756,598,193],"EmPqBF962_oISNcYZ9pLsSWouUcxswAx0nsnEA5ghp4",{"id":928,"title":929,"author":7,"body":930,"date":260,"description":1170,"difficulty":262,"extension":263,"image":264,"lastModified":260,"meta":1171,"navigation":266,"outcome":1172,"path":1173,"readingTime":674,"robots":264,"scenario":1174,"seo":1175,"stem":1176,"tools":1177,"__hash__":1179},"guides\u002F8.guides\u002F5.separate-vocals-from-a-track.md","How to Split a Track into Vocals and Instrumental",{"type":9,"value":931,"toc":1160},[932,935,939,942,956,959,962,966,972,978,981,985,993,999,1004,1007,1015,1021,1026,1030,1033,1036,1042,1048,1054,1060,1065,1069,1072,1075,1081,1086,1090,1126,1129,1131,1137,1143,1149,1155,1157],[12,933,934],{},"Separating a mixed track into vocals and instrumental sounds like a magic trick until you try it. The results are useful, and they are also noticeably imperfect. Knowing which is which before you start saves an afternoon of disappointment.",[19,936,938],{"id":937},"what-this-actually-does","What this actually does",[12,940,941],{},"The separator runs Kim Vocal 2, a model trained to estimate which parts of a mixed signal belong to a voice and which belong to everything else. It outputs two tracks:",[115,943,944,950],{},[118,945,946,949],{},[59,947,948],{},"Vocals"," — the estimated voice",[118,951,952,955],{},[59,953,954],{},"Instrumental"," — everything the model did not classify as voice",[12,957,958],{},"Both are WAV files. Nothing is uploaded; the model and runtime download once (about 64 MB) and then run locally.",[12,960,961],{},"What it is not: a lossless restoration. The tool's own documentation says separation may leave instruments in the vocal track and may alter the voice. Drums and cymbals in particular bleed into vocals, and heavily processed vocals come out thinner than the original.",[19,963,965],{"id":964},"before-you-start-two-limits-that-shape-the-workflow","Before you start: two limits that shape the workflow",[12,967,968,971],{},[59,969,970],{},"Ten minutes maximum per file."," A longer file is rejected. This is the reason the workflow has a trimming step at all — not tidiness, a hard limit.",[12,973,974,977],{},[59,975,976],{},"One file at a time, desktop only."," This is a desktop preview, not a batch tool. Feeding it a folder is not an option.",[12,979,980],{},"There is also a licensing point worth stating: process audio you own or are licensed to use. Splitting a commercial track does not give you rights to the separated parts.",[19,982,984],{"id":983},"step-1-trim-to-the-part-you-need","Step 1: Trim to the part you need",[12,986,987,988,992],{},"Open the source in the ",[48,989,991],{"href":990},"\u002Ftools\u002Faudio-trimmer","audio trimmer"," and cut a section under ten minutes.",[12,994,995],{},[75,996],{"alt":997,"src":998},"A long track being cut down to a section under the ten-minute separation limit","\u002Fimg\u002Fguides\u002Faudio-trim-to-limit.svg",[12,1000,1001],{},[64,1002,1003],{},"If the section you want is longer than ten minutes, split it into two and process each half. The seams rarely matter for what people use these tracks for.",[12,1005,1006],{},"Two habits help here:",[115,1008,1009,1012],{},[118,1010,1011],{},"Cut at a phrase boundary, not mid-word. The model works better on contiguous speech or a coherent musical phrase than on a fragment that starts halfway through a syllable.",[118,1013,1014],{},"Keep the trimmed file. Re-running separation on the same input gives slightly different results, and you will want a stable reference while comparing.",[12,1016,1017],{},[75,1018],{"alt":1019,"src":1020},"Speech waveform with start at one second, end at four seconds and a three-second selection","\u002Fimg\u002Fguides\u002Faudio-trimmer-range.jpg",[12,1022,1023],{},[64,1024,1025],{},"Generated speech sample with 0:01–0:04 selected. This short example demonstrates the controls without implying a ten-minute music test.",[19,1027,1029],{"id":1028},"step-2-preview-for-30-seconds-before-committing","Step 2: Preview for 30 seconds before committing",[12,1031,1032],{},"The separator offers a preview of the first 30 seconds. Use it. The full run on CPU can take several times the length of the audio, so finding out the input is bad after twenty minutes of processing is a waste.",[12,1034,1035],{},"In the preview, listen for three things:",[12,1037,1038,1041],{},[59,1039,1040],{},"Bleed."," Play the vocal track and listen for drums, bass or reverb tails. A dense mix leaks; a sparse one usually does not.",[12,1043,1044,1047],{},[59,1045,1046],{},"Thinning."," Play the instrumental and listen for whether the voice is really gone or just quiet. Residual vocal in the instrumental is the most common complaint, and it is normal rather than a bug.",[12,1049,1050,1053],{},[59,1051,1052],{},"Level."," The two tracks are exported with a shared output adjustment so they stay balanced against each other. If one sounds much louder in the preview, that is the mix, not the export.",[12,1055,1056],{},[75,1057],{"alt":1058,"src":1059},"Vocal separation result with original vocals and instrumental playback choices","\u002Fimg\u002Fguides\u002Fvocal-separator-preview.jpg",[12,1061,1062],{},[64,1063,1064],{},"Interface demonstration using generated speech rather than a mixed song. The separated outputs do not establish music separation quality.",[19,1066,1068],{"id":1067},"step-3-run-the-full-section-and-export","Step 3: Run the full section and export",[12,1070,1071],{},"Start the separation and let it finish. GPU acceleration helps when the browser exposes it; on integrated graphics the tool falls back to CPU and warns you that processing may take several times the audio length.",[12,1073,1074],{},"Results are held in temporary browser storage. A refresh can restore them for about ten minutes, but that is a convenience, not a save. Download both WAV files to your device as soon as the run completes.",[12,1076,1077],{},[75,1078],{"alt":1079,"src":1080},"Completed separation with separate vocals and instrumental WAV downloads","\u002Fimg\u002Fguides\u002Fvocal-separator-download.jpg",[12,1082,1083],{},[64,1084,1085],{},"Actual full-file processing of the short generated speech sample. Download both WAV outputs before leaving the workflow.",[19,1087,1089],{"id":1088},"what-people-do-with-the-two-tracks","What people do with the two tracks",[115,1091,1092,1098,1104,1115],{},[118,1093,1094,1097],{},[59,1095,1096],{},"Practice"," — play along to the instrumental, or sing over it",[118,1099,1100,1103],{},[59,1101,1102],{},"Cover or remix"," — with the instrumental as a backing bed",[118,1105,1106,1109,1110,1114],{},[59,1107,1108],{},"Transcript cleanup"," — send the vocal track to ",[48,1111,1113],{"href":1112},"\u002Ftools\u002Faudio-to-text","Audio to Text","; with less music competing, recognition is usually cleaner",[118,1116,1117,1120,1121,1125],{},[59,1118,1119],{},"Format conversion"," — WAV is large; use ",[48,1122,1124],{"href":1123},"\u002Ftools\u002Faudio-converter","Audio Converter"," to get MP3 or M4A for anything you intend to share",[12,1127,1128],{},"The transcript route is worth calling out because it is the one case where separation measurably improves another tool's output, rather than just producing a file you wanted.",[19,1130,428],{"id":427},[12,1132,1133,1136],{},[59,1134,1135],{},"The file is rejected as too long."," You skipped trimming, or the trim did not take because the selection was still over ten minutes. The limit counts the trimmed output, not the source.",[12,1138,1139,1142],{},[59,1140,1141],{},"The vocals sound underwater."," Aggressive separation artifacts. Try a different section of the same track; dense passages are harder than sparse ones.",[12,1144,1145,1148],{},[59,1146,1147],{},"The instrumental still has vocals in it."," Expected in most material. Lower the vocal track's contribution is not an option here — if you need a clean instrumental, this tool cannot produce one from a fully mixed master.",[12,1150,1151,1154],{},[59,1152,1153],{},"Processing never finishes on CPU."," Cancel rather than wait; nothing is lost, completed work is discarded and temporary files are released. Trim shorter, or try a machine with a usable GPU.",[19,1156,233],{"id":232},[12,1158,1159],{},"Two WAV files that are good enough for practice, backing tracks and transcript cleanup, and not good enough to pass as a studio stem. That distinction is the honest summary of what separation at this level does.",{"title":249,"searchDepth":250,"depth":250,"links":1161},[1162,1163,1164,1165,1166,1167,1168,1169],{"id":937,"depth":250,"text":938},{"id":964,"depth":250,"text":965},{"id":983,"depth":250,"text":984},{"id":1028,"depth":250,"text":1029},{"id":1067,"depth":250,"text":1068},{"id":1088,"depth":250,"text":1089},{"id":427,"depth":250,"text":428},{"id":232,"depth":250,"text":233},"The separator has a ten-minute ceiling, so trimming comes first. What to check in the 30-second preview, why the output is two WAV files, and what this kind of separation cannot do.",{},"Two WAV files — vocals and instrumental — from one source track you are allowed to process.","\u002Fguides\u002Fseparate-vocals-from-a-track","edit",{"title":929,"description":1170},"8.guides\u002F5.separate-vocals-from-a-track",[990,1178],"\u002Ftools\u002Fvocal-separator","2bmFHowDjWXjRcg0gSBGexCCu5PR1LnrCHwy9VLhDf8",{"id":1181,"title":1182,"author":7,"body":1183,"date":260,"description":1380,"difficulty":477,"extension":263,"image":264,"lastModified":260,"meta":1381,"navigation":266,"outcome":1382,"path":1383,"readingTime":674,"robots":264,"scenario":482,"seo":1384,"stem":1385,"tools":1386,"__hash__":1387},"guides\u002F8.guides\u002F6.pull-one-answer-from-a-podcast.md","How to Pull One Answer Out of a Two-Hour Podcast",{"type":9,"value":1184,"toc":1371},[1185,1188,1191,1195,1198,1201,1204,1208,1219,1225,1235,1241,1246,1249,1253,1256,1262,1265,1271,1276,1279,1284,1289,1293,1299,1302,1305,1311,1316,1320,1323,1337,1340,1342,1348,1354,1360,1366,1368],[12,1186,1187],{},"Someone said something useful on a podcast. You remember roughly where — first third, after the second ad break. You need the exact wording for a citation, a note, or a reply.",[12,1189,1190],{},"The obvious move is to transcribe the whole episode and search the text. That works, and it is the slowest possible path. This guide is about the faster one.",[19,1192,1194],{"id":1193},"why-transcribing-everything-is-the-wrong-default","Why transcribing everything is the wrong default",[12,1196,1197],{},"Recognition cost is proportional to duration. Two hours of audio takes roughly four times as long as thirty minutes and produces a file that is ninety percent irrelevant to what you need.",[12,1199,1200],{},"Podcasts also have a structure that helps you cut first: ad breaks, self-promotion, and repeated recaps. Those are the parts you would skip anyway.",[12,1202,1203],{},"The workflow that follows is: find the passage by listening at speed, cut it out, then transcribe eight minutes instead of two hours.",[19,1205,1207],{"id":1206},"step-1-get-the-audio-file","Step 1: Get the audio file",[12,1209,1210,1211,1214,1215,1218],{},"Find the audio source. On most podcast sites it is a direct link to an MP3 or M4A, and the easiest way to confirm is the Network panel filtered for ",[33,1212,1213],{},"media",", or looking at the page source for an ",[33,1216,1217],{},"\u003Caudio>"," element or an enclosure URL in the feed.",[12,1220,1221,1222,1224],{},"Paste the link into the ",[48,1223,222],{"href":221},". It accepts direct links to MP3, M4A\u002FAAC, WAV, OGG and FLAC.",[12,1226,1227,1228,1230,1231,1234],{},"If the podcast is delivered as an HLS audio playlist rather than a direct file — increasingly common for platform-hosted shows — use the ",[48,1229,137],{"href":136}," instead, and see the ",[48,1232,1233],{"href":268},"course archiving guide"," for how to read the playlist first.",[12,1236,1237],{},[75,1238],{"alt":1239,"src":1240},"Direct audio URL and ready-to-save WAV result with a file size of 180.5 KB","\u002Fimg\u002Fguides\u002Faudio-downloader-url.jpg",[12,1242,1243],{},[64,1244,1245],{},"Local public URL for the generated speech sample, fetched without the extension. This result panel shows size and format; use Inspect audio to check duration.",[12,1247,1248],{},"One confirmation worth making before you invest time: check that the file's duration matches the episode. A 12-minute file for a two-hour episode means you got a trailer or a preview clip, not the episode.",[19,1250,1252],{"id":1251},"step-2-find-the-passage-then-cut-it","Step 2: Find the passage, then cut it",[12,1254,1255],{},"This is the step people skip, and it is the one that saves the most time.",[12,1257,1258,1259,1261],{},"Open the downloaded file in the ",[48,1260,991],{"href":990},". Play at 1.5x or 2x if the tool allows it, or scrub with the waveform, and note the in and out points. Aim for a window of two to ten minutes that contains the whole argument — not just the sentence you want.",[12,1263,1264],{},"Cutting a little wide is deliberate. A transcript of three sentences tells you nothing about what was being answered, and you will end up going back to the audio anyway.",[12,1266,1267],{},[75,1268],{"alt":1269,"src":1270},"A two-hour podcast trimmed to an eight-minute window containing the passage, which is then transcribed instead of the whole episode","\u002Fimg\u002Fguides\u002Fpodcast-excerpt.svg",[12,1272,1273],{},[64,1274,1275],{},"Locate first, cut second, transcribe third. The order is what makes this take minutes instead of an hour.",[12,1277,1278],{},"While you are here, cut the ad reads out of the surrounding material. They carry no information and they are the part recognition handles worst, because music and jingles sit under the voice.",[12,1280,1281],{},[75,1282],{"alt":1283,"src":1020},"Audio waveform with a three-second selected excerpt and editable start and end times",[12,1285,1286],{},[64,1287,1288],{},"Interface example using generated speech, not a podcast recording. Select the eight-minute passage in your own episode using the same start and end controls.",[19,1290,1292],{"id":1291},"step-3-transcribe-the-clip","Step 3: Transcribe the clip",[12,1294,1295,1296,1298],{},"Send the clip to ",[48,1297,1113],{"href":1112},". Podcast audio is usually the easiest material this tool sees: one or two speakers, decent microphones, a consistent language.",[12,1300,1301],{},"Set the spoken language explicitly rather than relying on auto-detection. Detection samples the opening seconds, and a clip that starts mid-sentence has less to work with than a full episode would.",[12,1303,1304],{},"Choose the standard model if the passage contains names, numbers or technical terms — which, for a quotable passage, it usually does. The lightweight model is fine for confirming \"yes, this is the part I meant\".",[12,1306,1307],{},[75,1308],{"alt":1309,"src":1310},"Audio transcript with timestamps, editable text and subtitle export controls","\u002Fimg\u002Fguides\u002Faudio-to-text-clip.jpg",[12,1312,1313],{},[64,1314,1315],{},"Actual local transcription of the generated English speech sample using Whisper Tiny on CPU. This demonstrates the interface, not recognition quality on a podcast.",[19,1317,1319],{"id":1318},"step-4-check-the-quote-before-you-use-it","Step 4: Check the quote before you use it",[12,1321,1322],{},"Recognition gets the gist and misses the details that matter in a quotation. Before you attribute a sentence to someone:",[115,1324,1325,1328,1331,1334],{},[118,1326,1327],{},"Verify proper nouns by listening again at that timestamp",[118,1329,1330],{},"Check numbers — \"fifteen\" and \"fifty\" are a known weak point",[118,1332,1333],{},"Confirm who said it. This version does not identify speakers, so in a two-person interview you have to map lines to people yourself",[118,1335,1336],{},"Keep the timestamp in your notes. If the claim is ever disputed, the timestamp in the episode is your evidence",[12,1338,1339],{},"Transcript timestamps are relative to the clip you cut, not to the original episode. Add the offset from step 2 if you are citing a time in the full recording.",[19,1341,428],{"id":427},[12,1343,1344,1347],{},[59,1345,1346],{},"The download link 403s."," Podcast CDNs sign URLs, and the token in a copied link expires. Reload the episode page and copy a fresh link.",[12,1349,1350,1353],{},[59,1351,1352],{},"The clip starts mid-word."," The transcription begins with a fragment, and the model may misread it. Trim two seconds earlier than you think you need.",[12,1355,1356,1359],{},[59,1357,1358],{},"The transcript has the right words in the wrong order."," Long pauses and crosstalk confuse segmentation. For a short clip this is usually fixable by hand in a few seconds.",[12,1361,1362,1365],{},[59,1363,1364],{},"The episode is behind a login."," This workflow assumes the audio is publicly accessible. It is not a paywall workaround, and the tools do not attempt to be one.",[19,1367,233],{"id":232},[12,1369,1370],{},"An eight-minute clip and its transcript, both timestamped, plus a note of the offset into the original episode. The clip is also reusable if you need to check the audio again later, which a bare transcript is not.",{"title":249,"searchDepth":250,"depth":250,"links":1372},[1373,1374,1375,1376,1377,1378,1379],{"id":1193,"depth":250,"text":1194},{"id":1206,"depth":250,"text":1207},{"id":1251,"depth":250,"text":1252},{"id":1291,"depth":250,"text":1292},{"id":1318,"depth":250,"text":1319},{"id":427,"depth":250,"text":428},{"id":232,"depth":250,"text":233},"You do not need to transcribe two hours to find eight minutes. Locate the passage in the audio, cut it, then transcribe the cut. A guide to the shortest path from a podcast URL to a quotable transcript.",{},"An eight-minute audio clip and its transcript, instead of a two-hour one.","\u002Fguides\u002Fpull-one-answer-from-a-podcast",{"title":1182,"description":1380},"8.guides\u002F6.pull-one-answer-from-a-podcast",[221,990,1112],"2dhYZCyduUve5pAFGpOwBOCWWlAoGC5kEZCg4Wl0j_w",{"id":1389,"title":1390,"author":7,"body":1391,"date":260,"description":1625,"difficulty":477,"extension":263,"image":264,"lastModified":260,"meta":1626,"navigation":266,"outcome":1627,"path":1628,"readingTime":674,"robots":264,"scenario":1174,"seo":1629,"stem":1630,"tools":1631,"__hash__":1632},"guides\u002F8.guides\u002F7.clean-up-a-field-recording.md","How to Make a Noisy Field Recording Usable",{"type":9,"value":1392,"toc":1616},[1393,1396,1399,1403,1406,1412,1418,1424,1430,1433,1437,1440,1448,1454,1459,1462,1466,1469,1475,1480,1483,1490,1493,1499,1505,1511,1516,1519,1523,1530,1533,1541,1547,1552,1555,1559,1562,1570,1576,1582,1584,1590,1596,1602,1611,1613],[12,1394,1395],{},"A recording made in a real room sounds different from the one you imagined. There is a fan, a fridge, traffic, and the speaker was three metres from the microphone. Usually both problems appear together: too quiet and too noisy.",[12,1397,1398],{},"The instinct is to turn the volume up first, because that is the obvious problem. That is the wrong order, and the rest of this guide explains why.",[19,1400,1402],{"id":1401},"what-noise-removal-can-and-cannot-do","What noise removal can and cannot do",[12,1404,1405],{},"The denoiser runs RNNoise, which processes voice in short frames and estimates what is noise versus speech. That design sets its limits:",[12,1407,1408,1411],{},[59,1409,1410],{},"It handles steady noise well."," Fan hum, air conditioning, room tone, a distant motorway — anything with a stable spectrum.",[12,1413,1414,1417],{},[59,1415,1416],{},"It handles other voices poorly."," If a second person talked in the background, they are speech, not noise. The model will keep them and may mangle them.",[12,1419,1420,1423],{},[59,1421,1422],{},"It cannot recover masked words."," Noise that completely covers a syllable means that syllable is gone. No processing brings it back.",[12,1425,1426,1429],{},[59,1427,1428],{},"It can damage music."," If your recording contains music, expect it to sound odd afterwards. This is a voice tool.",[12,1431,1432],{},"Knowing which of these you have decides whether this workflow is worth starting.",[19,1434,1436],{"id":1435},"step-1-get-the-audio-out-of-whatever-it-is-in","Step 1: Get the audio out of whatever it is in",[12,1438,1439],{},"Field recordings usually arrive inside a video container: a phone video, a screen recording, a camera file.",[12,1441,1442,1443,1447],{},"Use the ",[48,1444,1446],{"href":1445},"\u002Ftools\u002Faudio-extractor","audio extractor"," to pull the audio track out. Working on a 40 MB audio file instead of a 2 GB video is faster at every later step, and it makes the intermediates easier to keep organised.",[12,1449,1450],{},[75,1451],{"alt":1452,"src":1453},"Extracted M4A audio result showing AAC codec, original quality, no re-encoding and 33.1 KB output","\u002Fimg\u002Fguides\u002Faudio-extractor-output.jpg",[12,1455,1456],{},[64,1457,1458],{},"Generated four-second MP4 → 33.1 KB AAC\u002FM4A audio. Original audio mode copies the audio packets without re-encoding.",[12,1460,1461],{},"If the source is already an audio file, skip this step.",[19,1463,1465],{"id":1464},"step-2-denoise-before-you-turn-anything-up","Step 2: Denoise before you turn anything up",[12,1467,1468],{},"This is the part that decides the outcome.",[12,1470,1471],{},[75,1472],{"alt":1473,"src":1474},"Two paths: amplifying first raises the noise floor with the voice, while denoising first then amplifying produces a cleaner result","\u002Fimg\u002Fguides\u002Fdenoise-before-gain.svg",[12,1476,1477],{},[64,1478,1479],{},"Gain applied first is applied to the noise too, and the denoiser then has a harder job.",[12,1481,1482],{},"Turning the volume up first multiplies everything, noise included. The denoiser then sees a louder but equally noisy signal, and its estimate is no better — but your ears now notice the artefacts more.",[12,1484,302,1485,1489],{},[48,1486,1488],{"href":1487},"\u002Ftools\u002Faudio-noise-remover","noise remover"," and process the original level.",[12,1491,1492],{},"Two settings matter:",[12,1494,1495,1498],{},[59,1496,1497],{},"Preview first."," The tool processes the first 30 seconds as a preview. Always listen to that before running the full recording, because the full run on a long file takes time and the result is not always better.",[12,1500,1501,1504],{},[59,1502,1503],{},"Denoise amount."," The output mix is a percentage, and lower values blend more of the original back in. Full strength removes the most noise and also the most character — voices start to sound processed. For material you intend to publish, somewhere below full strength usually sounds better; for material you only intend to transcribe, push it higher.",[12,1506,1507],{},[75,1508],{"alt":1509,"src":1510},"RNNoise interface showing 80 percent denoised mix, original and denoised playback, and WAV export","\u002Fimg\u002Fguides\u002Fnoise-remover-compare.jpg",[12,1512,1513],{},[64,1514,1515],{},"Interface demonstration with generated speech at an 80% mix. This is not a noisy field recording or evidence of improved speech quality.",[12,1517,1518],{},"The tool's limit is 30 minutes per file. Longer recordings need to be split, which is also a reasonable excuse to cut the parts you will not use.",[19,1520,1522],{"id":1521},"step-3-raise-the-level-gently","Step 3: Raise the level, gently",[12,1524,1525,1526,138],{},"Now the recording is clean but still quiet. Open it in the ",[48,1527,1529],{"href":1528},"\u002Ftools\u002Faudio-volume-booster","volume booster",[12,1531,1532],{},"Raise until it is comfortable, not until it is loud. Two reasons:",[115,1534,1535,1538],{},[118,1536,1537],{},"Boosting also raises whatever noise survived, and after denoising that residue is often hiss the model could not separate from breath.",[118,1539,1540],{},"If the recording clips — peaks hitting the ceiling — the damage is permanent. Leave headroom.",[12,1542,1543],{},[75,1544],{"alt":1545,"src":1546},"Volume booster with plus three dB gain, peak protection and original versus processed playback","\u002Fimg\u002Fguides\u002Fvolume-booster-gain.jpg",[12,1548,1549],{},[64,1550,1551],{},"Generated speech with +3 dB gain and peak protection enabled. The displayed 61% peak is the source analysis; the interface does not display a measured output peak.",[12,1553,1554],{},"Listen for the sibilance: \"s\" and \"sh\" sounds are the first thing to distort when gain goes too high.",[19,1556,1558],{"id":1557},"step-4-decide-what-the-file-is-for","Step 4: Decide what the file is for",[12,1560,1561],{},"The same recording needs different treatment depending on the destination, and it is worth being explicit about which one you want before exporting.",[12,1563,1564,1567,1568,138],{},[59,1565,1566],{},"For transcription."," Denoise hard, keep the level moderate. Recognition is more sensitive to the noise floor than to absolute loudness. Send the result to ",[48,1569,1113],{"href":1112},[12,1571,1572,1575],{},[59,1573,1574],{},"For sharing."," Denoise gently and normalise the level. Over-processed speech sounds worse to a listener than mild background noise does.",[12,1577,1578,1581],{},[59,1579,1580],{},"For editing."," Export a WAV at the processed level and let the editor do the final level matching. Do not apply the last 3 dB here.",[19,1583,428],{"id":427},[12,1585,1586,1589],{},[59,1587,1588],{},"The result sounds robotic."," Too much denoise. Go back to the original and reprocess at a lower amount; the processing is not incremental, so re-running from the source is required.",[12,1591,1592,1595],{},[59,1593,1594],{},"Noise is still there after processing."," It was probably not steady noise, or it was too loud relative to the voice. Check whether it is a fan (should work) or a conversation (will not work).",[12,1597,1598,1601],{},[59,1599,1600],{},"The voice sounds distant and thin."," The microphone was far away. Denoising cannot add proximity; it can only remove what is not voice, and a distant voice has little of what makes speech sound present.",[12,1603,1604,1607,1608,1610],{},[59,1605,1606],{},"The file got bigger."," Denoised output is WAV regardless of the input format. Convert to MP3 or M4A with ",[48,1609,1124],{"href":1123}," if size matters.",[19,1612,233],{"id":232},[12,1614,1615],{},"A recording that is usable for its purpose, and an honest sense of what that purpose can be. A phone recording from across a room will not become a studio interview, but it can become a perfectly searchable transcript, which is often the actual goal.",{"title":249,"searchDepth":250,"depth":250,"links":1617},[1618,1619,1620,1621,1622,1623,1624],{"id":1401,"depth":250,"text":1402},{"id":1435,"depth":250,"text":1436},{"id":1464,"depth":250,"text":1465},{"id":1521,"depth":250,"text":1522},{"id":1557,"depth":250,"text":1558},{"id":427,"depth":250,"text":428},{"id":232,"depth":250,"text":233},"Recorded on a phone in a room with a fan? Noise removal only helps with certain kinds of noise, and the order of denoise and gain matters more than either setting. A practical sequence.",{},"A voice recording that is loud enough and clean enough to transcribe or publish.","\u002Fguides\u002Fclean-up-a-field-recording",{"title":1390,"description":1625},"8.guides\u002F7.clean-up-a-field-recording",[1445,1487,1528],"1PsuwkWjHZ0y1-R9DoYV-z5L0x6RUJxRSyf0EDN6TT8",1793546706294]