ClosedCaptionDecoder

Turns the closed captions that video frames carry into timed text (#179): the bytes Frame.closedCaptions gives, three per caption pair, as broadcast H.264, HEVC and MPEG-2 carry them in the video stream itself. A caption track stored on its own, such as a MOV c608 track, is a subtitle stream instead and decodes through MediaSource.openSubtitleDecoder.

This is FFmpeg's own caption decoder, eia_608, the one mpv feeds each frame's captions to. It reads the CEA-608 captions of field 1, CC1 and CC2, which is where broadcast puts its captions, those inside CEA-708 data included, and skips the CEA-708 service blocks, which FFmpeg does not decode. The field is fixed rather than guessed: left to guess, FFmpeg's decoder takes the field of the first byte it is given, before it checks that byte, so one damaged or CEA-708 byte at the start leaves it reading a field nothing is captioned in.

Hand over each frame's bytes in the order the frames are shown, which is the order the captions were authored in and not the order a decoder with reordered pictures receives them. Each answer is a Subtitle in the text form SubtitleDecoder gives, one ASS event per rectangle, timed on the caller's timeline: the times given to decode.

The decoder answers in one of two ways, chosen at open. Buffered, the default, gives each caption once it leaves the screen, with its start and its end, because only then is its end known; the one on screen when the bytes run out comes from drain. That suits a converter writing a subtitle file. Real time gives the screen as it now stands each time it changes, from the time of the frame that changed it, with no end: each answer holds until the next one, and an answer whose text is empty clears the screen. Roll-up and paint-on captions answer as they are painted, at most once every 200 ms. That suits a player, which decodes pictures only a little ahead of the one on screen and would otherwise get each caption as it disappears; mpv opens FFmpeg's decoder this way for every caption stream it plays (#180).

Not thread safe: use one decoder from one thread at a time.

Turns the closed captions that video frames carry into timed text (#179): the bytes Frame.closedCaptions gives, three per caption pair, as broadcast H.264, HEVC and MPEG-2 carry them in the video stream itself. A caption track stored on its own, such as a MOV c608 track, is a subtitle stream instead and decodes through MediaSource.openSubtitleDecoder.

This is FFmpeg's own caption decoder, eia_608, the one mpv feeds each frame's captions to. It reads the CEA-608 captions of field 1, CC1 and CC2, which is where broadcast puts its captions, those inside CEA-708 data included, and skips the CEA-708 service blocks, which FFmpeg does not decode. The field is fixed rather than guessed: left to guess, FFmpeg's decoder takes the field of the first byte it is given, before it checks that byte, so one damaged or CEA-708 byte at the start leaves it reading a field nothing is captioned in.

Hand over each frame's bytes in the order the frames are shown, which is the order the captions were authored in and not the order a decoder with reordered pictures receives them. Each answer is a Subtitle in the text form SubtitleDecoder gives, one ASS event per rectangle, timed on the caller's timeline: the times given to decode.

The decoder answers in one of two ways, chosen at open. Buffered, the default, gives each caption once it leaves the screen, with its start and its end, because only then is its end known; the one on screen when the bytes run out comes from drain. That suits a converter writing a subtitle file. Real time gives the screen as it now stands each time it changes, from the time of the frame that changed it, with no end: each answer holds until the next one, and an answer whose text is empty clears the screen. Roll-up and paint-on captions answer as they are painted, at most once every 200 ms. That suits a player, which decodes pictures only a little ahead of the one on screen and would otherwise get each caption as it disappears; mpv opens FFmpeg's decoder this way for every caption stream it plays (#180).

Not thread safe: use one decoder from one thread at a time.

Turns the closed captions that video frames carry into timed text (#179): the bytes Frame.closedCaptions gives, three per caption pair, as broadcast H.264, HEVC and MPEG-2 carry them in the video stream itself. A caption track stored on its own, such as a MOV c608 track, is a subtitle stream instead and decodes through MediaSource.openSubtitleDecoder.

This is FFmpeg's own caption decoder, eia_608, the one mpv feeds each frame's captions to. It reads the CEA-608 captions of field 1, CC1 and CC2, which is where broadcast puts its captions, those inside CEA-708 data included, and skips the CEA-708 service blocks, which FFmpeg does not decode. The field is fixed rather than guessed: left to guess, FFmpeg's decoder takes the field of the first byte it is given, before it checks that byte, so one damaged or CEA-708 byte at the start leaves it reading a field nothing is captioned in.

Hand over each frame's bytes in the order the frames are shown, which is the order the captions were authored in and not the order a decoder with reordered pictures receives them. Each answer is a Subtitle in the text form SubtitleDecoder gives, one ASS event per rectangle, timed on the caller's timeline: the times given to decode.

The decoder answers in one of two ways, chosen at open. Buffered, the default, gives each caption once it leaves the screen, with its start and its end, because only then is its end known; the one on screen when the bytes run out comes from drain. That suits a converter writing a subtitle file. Real time gives the screen as it now stands each time it changes, from the time of the frame that changed it, with no end: each answer holds until the next one, and an answer whose text is empty clears the screen. Roll-up and paint-on captions answer as they are painted, at most once every 200 ms. That suits a player, which decodes pictures only a little ahead of the one on screen and would otherwise get each caption as it disappears; mpv opens FFmpeg's decoder this way for every caption stream it plays (#180).

Not thread safe: use one decoder from one thread at a time.

Turns the closed captions that video frames carry into timed text (#179): the bytes Frame.closedCaptions gives, three per caption pair, as broadcast H.264, HEVC and MPEG-2 carry them in the video stream itself. A caption track stored on its own, such as a MOV c608 track, is a subtitle stream instead and decodes through MediaSource.openSubtitleDecoder.

This is FFmpeg's own caption decoder, eia_608, the one mpv feeds each frame's captions to. It reads the CEA-608 captions of field 1, CC1 and CC2, which is where broadcast puts its captions, those inside CEA-708 data included, and skips the CEA-708 service blocks, which FFmpeg does not decode. The field is fixed rather than guessed: left to guess, FFmpeg's decoder takes the field of the first byte it is given, before it checks that byte, so one damaged or CEA-708 byte at the start leaves it reading a field nothing is captioned in.

Hand over each frame's bytes in the order the frames are shown, which is the order the captions were authored in and not the order a decoder with reordered pictures receives them. Each answer is a Subtitle in the text form SubtitleDecoder gives, one ASS event per rectangle, timed on the caller's timeline: the times given to decode.

The decoder answers in one of two ways, chosen at open. Buffered, the default, gives each caption once it leaves the screen, with its start and its end, because only then is its end known; the one on screen when the bytes run out comes from drain. That suits a converter writing a subtitle file. Real time gives the screen as it now stands each time it changes, from the time of the frame that changed it, with no end: each answer holds until the next one, and an answer whose text is empty clears the screen. Roll-up and paint-on captions answer as they are painted, at most once every 200 ms. That suits a player, which decodes pictures only a little ahead of the one on screen and would otherwise get each caption as it disappears; mpv opens FFmpeg's decoder this way for every caption stream it plays (#180).

Not thread safe: use one decoder from one thread at a time.

Turns the closed captions that video frames carry into timed text (#179): the bytes Frame.closedCaptions gives, three per caption pair, as broadcast H.264, HEVC and MPEG-2 carry them in the video stream itself. A caption track stored on its own, such as a MOV c608 track, is a subtitle stream instead and decodes through MediaSource.openSubtitleDecoder.

This is FFmpeg's own caption decoder, eia_608, the one mpv feeds each frame's captions to. It reads the CEA-608 captions of field 1, CC1 and CC2, which is where broadcast puts its captions, those inside CEA-708 data included, and skips the CEA-708 service blocks, which FFmpeg does not decode. The field is fixed rather than guessed: left to guess, FFmpeg's decoder takes the field of the first byte it is given, before it checks that byte, so one damaged or CEA-708 byte at the start leaves it reading a field nothing is captioned in.

Hand over each frame's bytes in the order the frames are shown, which is the order the captions were authored in and not the order a decoder with reordered pictures receives them. Each answer is a Subtitle in the text form SubtitleDecoder gives, one ASS event per rectangle, timed on the caller's timeline: the times given to decode.

The decoder answers in one of two ways, chosen at open. Buffered, the default, gives each caption once it leaves the screen, with its start and its end, because only then is its end known; the one on screen when the bytes run out comes from drain. That suits a converter writing a subtitle file. Real time gives the screen as it now stands each time it changes, from the time of the frame that changed it, with no end: each answer holds until the next one, and an answer whose text is empty clears the screen. Roll-up and paint-on captions answer as they are painted, at most once every 200 ms. That suits a player, which decodes pictures only a little ahead of the one on screen and would otherwise get each caption as it disappears; mpv opens FFmpeg's decoder this way for every caption stream it plays (#180).

Not thread safe: use one decoder from one thread at a time.