Open-source X Spaces transcription.

Star

87 lines · 2.6 KB

json
1[
2 {
3 "start": 0.8,
4 "end": 4.99,
5 "speaker": "tape_op",
6 "text": "This is what a Space looks like after space-tape runs."
7 },
8 {
9 "start": 5.1,
10 "end": 13.08,
11 "speaker": "tape_op",
12 "text": "A replay is just a file. You do not need a browser, and you do not need an X API key."
13 },
14 {
15 "start": 13.08,
16 "end": 21.52,
17 "speaker": "second_mic",
18 "text": "Whisper heard the words. Hydra tagged who was unmuted. About once a second, X writes that into the audio as ID3."
19 },
20 {
21 "start": 21.52,
22 "end": 29.66,
23 "speaker": "tape_op",
24 "text": "HydraAudioLevel slot zero is the host. Everyone else on stage sits in HydraParticipants."
25 },
26 {
27 "start": 29.66,
28 "end": 37.99,
29 "speaker": "second_mic",
30 "text": "If you remux to wav before you parse, those tags are gone. The transcript still has words. It has no speakers."
31 },
32 {
33 "start": 37.99,
34 "end": 49.03,
35 "speaker": "tape_op",
36 "text": "So the pipeline is: yt-dlp, keep the mpegts, parse Hydra, then convert a copy for Whisper."
37 },
38 {
39 "start": 49.03,
40 "end": 57.1,
41 "speaker": "second_mic",
42 "text": "Each Whisper cue gets the Hydra region covering its midpoint. That is the whole diarizer. We are not guessing voices."
43 },
44 {
45 "start": 57.1,
46 "end": 66.93,
47 "speaker": "tape_op",
48 "text": "Overlaps stay labeled as both. The mix is one channel. Do not pretend Whisper can split two people talking at once."
49 },
50 {
51 "start": 66.93,
52 "end": 77.26,
53 "speaker": "second_mic",
54 "text": "The product is cues.json. Start, end, speaker, text. Plus transcript.md and a small mp3 for a player."
55 },
56 {
57 "start": 77.26,
58 "end": 85.51,
59 "speaker": "tape_op",
60 "text": "Always keep the original X URL. Playback on a site is a convenience. X is the source."
61 },
62 {
63 "start": 85.51,
64 "end": 94.0,
65 "speaker": "second_mic",
66 "text": "Public recorded Spaces only. Replays expire, so the download is the archive. Listeners never appear in the tags."
67 },
68 {
69 "start": 94.0,
70 "end": 105.63,
71 "speaker": "tape_op",
72 "text": "If yt-dlp finds no media, stop. The Space was not recorded, or the replay is gone. This tool does not join live."
73 },
74 {
75 "start": 105.63,
76 "end": 114.05,
77 "speaker": "second_mic",
78 "text": "Drop a raw replay on the player page to parse Hydra in the browser. Nothing is uploaded. The file already had the speaker list."
79 },
80 {
81 "start": 114.05,
82 "end": 120.81,
83 "speaker": "tape_op",
84 "text": "That is space-tape. Fork it. Run it on a shell. Host the git as a page."
85 }
86]