A film on this machine

A video does not have to be on YouTube. A lesson you recorded, a film you own, a course that came on a disk: give the add page the path to the file instead of an address, and it becomes a video like any other — the same glossing, the same clouds, the same cards — that plays with no internet at all. The file may be a sound as well as a film (below), and it may be sent from another device instead of named (below).

1. Adding one#

On the add page, answer Where is the video? with A video or a sound on this machine (Adding a video). Step 1 then asks for The video or sound: the full path of the file, as your file manager shows it — /home/you/films/lesson-1.mp4, C:\Users\you\Videos\lesson 1.mp4. Quotes around it are ignored, and ~ means your home folder. When you leave the box the page says what the file is: A video (mp4, 412 MB, 1:02:11). or A sound with no picture (mp3, 4.2 MB, 3:12): the player shows a bar with its waveform in place of a frame.

The page names .mp4, .webm, .mkv, .mov and .m4v; the server takes .avi, .ogv and .ogg as well, and the sounds below. Whether your browser can play the file is another matter — .mp4 and .webm play everywhere, .mkv and .avi often do not — and the player says so if it cannot (below).

What it refuses:

Either answer to Who writes the glosses? takes a film or a sound: the prompt for an LLM, or a video started empty to gloss in the player.

2. A sound instead of a film#

A recording that was never filmed is a video with no picture. Any of .mp3, .m4a, .aac, .ogg, .oga, .opus, .wav, .flac, .weba, .wma, .aiff, .aif, .amr, .mka and .caf is taken wherever a film is: by its path, sent, started empty, glossed by an LLM, transcribed by speech to text. What is different, and why:

Its kind is decided once.

video.json says "kind": "audio" and nothing else needs to look (a film’s video.json says nothing, as it always did). Where ffmpeg is installed it is asked what is inside the file, so an mp3 with the album’s picture in it is still a sound, and an MP4 with no picture in it is one too; without ffmpeg the extension decides.

A bar stands where the frame was.

The player draws the shape of the sound, whole, with the playing place on it — press anywhere on it to go there, drag to scrub — the browser’s own play controls under it, and a hairline at the foot where each caption starts. It is drawn by ffmpeg from the file; without ffmpeg the bar is a plain track and says so. The transcript follows the sound as it follows a film, and everything that worked on a film’s sound — the timings, a card’s recording — works on a sound’s.

A card has no frame.

The sheet’s frame row is not drawn, and the card takes the recording cut out of the sound (Adding a video says what the speech to text does with one).

The shelf says so.

A card of a sound carries ♪ a sound, where a thumbnail would be.

The prompt says “a recording”.

A prompt to an LLM names the video a recording on the reader’s own machine, not on YouTube.

A sound the browser cannot play#

.wma, .aiff, .aif, .amr, .mka and .caf — and a file whose sound is in a form a browser has no player for — get a playable copy, made by ffmpeg when the video is added (an mp3 where ffmpeg can write one, else an m4a or a wav). The copy is the video’s media.mp3; the original is kept beside it as media-orig.wma, hard-linked where the disk allows it, so it costs no space. The page says so when you leave the path box — .wma is a sound this browser cannot play as it is: a playable copy is made when the video is added, and the original stays beside it — and the answer says it again when it is done. A bundle carries the copy and not the original.

Where ffmpeg is not installed the sound is added as it is and the page says so, in words, before you go on: …ffmpeg, which would make a playable copy, is not installed on this computer: it is added as it is, and may stay silent. The player says this browser cannot play that sound if it is. Install ffmpeg and add the sound again.

3. Sending the file instead of naming it#

The path stays what it always was: the way to name a file that is on the computer, hard-linked, so a two-hour film costs no copy. Under the box there is a second way, Choose a video or a sound…, for a file that is on another device — a phone’s recording, a laptop’s film — or for a person on Windows who has no path to type:

1. choose the file. Before anything is sent the page says what it is — its size, whether it is a sound or a video, and how much room this computer’s disk has — and refuses in words what cannot be taken: name the file with its extension, that file is empty, there is no room on this computer’s disk for it: it needs 476 MB and 120 MB is free;
2. press Send it. A bar and a line say how far it has got, how fast, and about how long is left (Sending lesson.mp3: 38 % of 412 MB · 12 MB/s · about 22 seconds left); Stop ends it and nothing is kept;
3. when the last byte is in, the computer looks at the file — where ffmpeg is installed, a file that is not a video or a sound it can read is refused then — and puts its path in the box. From there it is a path like any other: the prompt, speech to text, Start it empty.

There is no limit on the size but the disk’s. A file is written whole or not at all — to a .part name nothing reads as media, and renamed only when it is all there — and what you send and never use is cleared after two days (it waits in youtube/videos/.incoming/, which is never listed, served or bundled). Adding the video links the file into the video’s folder and the waiting copy goes.

4. Its transcript: a panel, or a subtitle file#

The transcript box takes the same panel as for a YouTube video — or a whole .srt or .vtt subtitle file, pasted as it is. It is translated into the panel everything else here already reads — the .srt on the left becomes the panel on the right:

1                                      0:01.500:00:01,500 --> 00:00:04,000          Buongiorno, vorrei un chilo di pomodoriBuongiorno, vorrei un chilo            0:04.2di pomodori                            Quanto costano le mele oggi200:00:04,200 --> 00:00:07,000<i>Quanto costano le mele oggi</i>

• the cue numbers go, and so do tags such as <i> and the word timings of YouTube’s automatic captions;
• the lines of a cue become one caption, starting at the cue’s own time, fraction and all;
• YouTube’s rolling captions are collapsed: automatic captions repeat the line before with a few words added. A cue that begins with the whole of the caption before it keeps only what it adds, whatever the timing — nobody writes a subtitle that way on purpose. A cue that repeats the one before exactly is dropped when the two overlap in time, and kept when they do not: the same words later are somebody saying them again — a refrain, a correction;
• a caption that is only digits (1979) stays a caption;
• cues out of order are put in order.

A file with no cue it could read is refused: that subtitle file holds no cues this could read.

5. Or its transcript, made by speech to text#

A film with no subtitles can have its transcript made on this computer if speech to text is set up (Speech to text, in Settings; it is optional, and a pasted transcript or a subtitle file needs none of it). With the film’s path in step 1, the block under the transcript box offers Transcribe: Adding a video says what it asks and does.

For a film it is the simple case:

• nothing is played. The computer opens the file and listens to it itself, so the film’s format does not matter to your browser, and neither does the browser: any browser on any device that has been let in will do, a phone’s included;
• the path is a path on the computer’s own disk. When you are using the page from another device, name the file where the computer keeps it, not where you are;
• the transcript goes into the box, timed from the film’s own start, ready to be corrected in the editor — and the film is not added until you go on;
• the film is read where it is, never copied, and the audio is processed on this computer;
• no waveform is written: as always, a film’s is drawn from the film itself when the timings sheet wants it.

6. What happens to the film#

The film goes beside the transcript, in the video’s own folder, as media.mp4 (or whatever its extension is). It is hardlinked: a second name for the same file, which costs no disk and no time even for a two-hour film. Where the file system forbids a link — the film is on another disk — it is copied instead, which does take time and space; on the way to the player, a video started empty says which it was (The film was copied (1840 MB) — this filesystem would not take a link.).

Either way what lands is a real file, and the file being there is the whole of what makes the video local: nothing in video.json declares it, so there is no field to keep true. (A sound is the one thing video.json does say, "kind": "audio", because a file’s name does not always show whether it has a picture.) What follows from there being no YouTube:

• the video’s id is made from the file’s name — or, when you start it empty, from the title you typed — with six random hexadecimal characters after it (lesson-1-3fa2c1), so two lessons both called lesson 1 are two videos;
• video.json’s url is empty: no address is invented;
• the title is the file’s name when you give none, and the channel is on this machine;
• nothing is fetched from the internet, at any step.

The film is kept out of git, as a book’s recordings are — hundreds of megabytes are not what a repository is for — while the glosses beside it are ordinary files that git keeps. The film travels by the download button instead.

7. Playing it#

The player shows the browser’s own video, with its own controls, where a YouTube video’s frame would be — for a sound, the bar above — and everything else — the lit line, keep in view, Pause while a gloss is open, a click to replay, the clouds — works exactly the same. The film is served by Parseh itself, a piece at a time, which is what lets it seek anywhere at once. On a phone the bar fills the same place, held upright or sideways; a sound is never put beside the text, because there is no picture to put there.

When the film will not play#

The place of the video says which of these went wrong, and the transcript still works in every case:

8. What else a film makes possible#

• The timings draw its waveform from the film itself, on the server, with nothing to record first (The timings). This needs ffmpeg, as a book’s narration does.
• 🔊 cut the audio… on a card cuts the word’s own sound out of the film, on the server — no screen sharing, any browser. (The section Cards and Anki explains the card sheet.)
• A card’s link back to its source points at the player on this machine, at the card’s moment, rather than at youtube.com.

9. Taking it away with its film#

The ⤓ in the player’s bar hands over the film and its glosses in one zip, so that the video comes out of a bundle on another machine and plays there — a sound as a film does, as media.mp3 (its playable copy where one was made; the original stays here). An older Parseh that does not know the sound types takes the bundle all the same, leaves the sound out and says which file it left. Its tooltip says how big that zip is — (about 1.9 GB) — because a browser starts a two-hour film’s download without a word. For the words alone there is no button: add ?media=text to the download’s address, as the tooltip says. A book’s download defaults the other way, and the asymmetry is the point: a book without its recording is still the book, while a video without its film is a transcript of something nobody can watch. See Taking videos away and bringing them back.