Rewrite transcript logic to be more generic (#4747)

The transcript logic in Invidious was written specifically as a workaround for
captions, and not transcripts as a feature.

This PR genericises the logic as so it can be used to implement transcripts
within Invidious.

The most notable change is the added parsing of section headings when it was
previously skipped over in favor of regular lines.

src/invidious/routes/api/v1/videos.cr

@@ -89,9 +89,14 @@ module Invidious::Routes::API::V1::Videos
 
     if CONFIG.use_innertube_for_captions
       params = Invidious::Videos::Transcript.generate_param(id, caption.language_code, caption.auto_generated)
-      initial_data = YoutubeAPI.get_transcript(params)
 
-      webvtt = Invidious::Videos::Transcript.convert_transcripts_to_vtt(initial_data, caption.language_code)
+      transcript = Invidious::Videos::Transcript.from_raw(
+        YoutubeAPI.get_transcript(params),
+        caption.language_code,
+        caption.auto_generated
+      )
+
+      webvtt = transcript.to_vtt
     else
       # Timedtext API handling
       url = URI.parse("#{caption.base_url}&tlang=#{tlang}").request_target

src/invidious/videos/transcript.cr

@@ -1,8 +1,26 @@
 module Invidious::Videos
-  # Namespace for methods primarily relating to Transcripts
-  module Transcript
-    record TranscriptLine, start_ms : Time::Span, end_ms : Time::Span, line : String
+  # A `Transcripts` struct encapsulates a sequence of lines that together forms the whole transcript for a given YouTube video.
+  # These lines can be categorized into two types: section headings and regular lines representing content from the video.
+  struct Transcript
+    # Types
+    record HeadingLine, start_ms : Time::Span, end_ms : Time::Span, line : String
+    record RegularLine, start_ms : Time::Span, end_ms : Time::Span, line : String
+    alias TranscriptLine = HeadingLine | RegularLine
 
+    property lines : Array(TranscriptLine)
+
+    property language_code : String
+    property auto_generated : Bool
+
+    # User friendly label for the current transcript.
+    # Example: "English (auto-generated)"
+    property label : String
+
+    # Initializes a new Transcript struct with the contents and associated metadata describing it
+    def initialize(@lines : Array(TranscriptLine), @language_code : String, @auto_generated : Bool, @label : String)
+    end
+
+    # Generates a protobuf string to fetch the requested transcript from YouTube
     def self.generate_param(video_id : String, language_code : String, auto_generated : Bool) : String
       kind = auto_generated ? "asr" : ""
 
@@ -30,48 +48,79 @@ module Invidious::Videos
       return params
     end
 
-    def self.convert_transcripts_to_vtt(initial_data : Hash(String, JSON::Any), target_language : String) : String
-      # Convert into array of TranscriptLine
-      lines = self.parse(initial_data)
+    # Constructs a Transcripts struct from the initial YouTube response
+    def self.from_raw(initial_data : Hash(String, JSON::Any), language_code : String, auto_generated : Bool)
+      transcript_panel = initial_data.dig("actions", 0, "updateEngagementPanelAction", "content", "transcriptRenderer",
+        "content", "transcriptSearchPanelRenderer")
 
+      segment_list = transcript_panel.dig("body", "transcriptSegmentListRenderer")
+
+      if !segment_list["initialSegments"]?
+        raise NotFoundException.new("Requested transcript does not exist")
+      end
+
+      # Extract user-friendly label for the current transcript
+
+      footer_language_menu = transcript_panel.dig?(
+        "footer", "transcriptFooterRenderer", "languageMenu", "sortFilterSubMenuRenderer", "subMenuItems"
+      )
+
+      if footer_language_menu
+        label = footer_language_menu.as_a.select(&.["selected"].as_bool)[0]["title"].as_s
+      else
+        label = language_code
+      end
+
+      # Extract transcript lines
+
+      initial_segments = segment_list["initialSegments"].as_a
+
+      lines = [] of TranscriptLine
+
+      initial_segments.each do |line|
+        if unpacked_line = line["transcriptSectionHeaderRenderer"]?
+          line_type = HeadingLine
+        else
+          unpacked_line = line["transcriptSegmentRenderer"]
+          line_type = RegularLine
+        end
+
+        start_ms = unpacked_line["startMs"].as_s.to_i.millisecond
+        end_ms = unpacked_line["endMs"].as_s.to_i.millisecond
+        text = extract_text(unpacked_line["snippet"]) || ""
+
+        lines << line_type.new(start_ms, end_ms, text)
+      end
+
+      return Transcript.new(
+        lines: lines,
+        language_code: language_code,
+        auto_generated: auto_generated,
+        label: label
+      )
+    end
+
+    # Converts transcript lines to a WebVTT file
+    #
+    # This is used within Invidious to replace subtitles
+    # as to workaround YouTube's rate-limited timedtext endpoint.
+    def to_vtt
       settings_field = {
         "Kind"     => "captions",
-        "Language" => target_language,
+        "Language" => @language_code,
       }
 
-      # Taken from Invidious::Videos::Captions::Metadata.timedtext_to_vtt()
       vtt = WebVTT.build(settings_field) do |vtt|
-        lines.each do |line|
+        @lines.each do |line|
+          # Section headers are excluded from the VTT conversion as to
+          # match the regular captions returned from YouTube as much as possible
+          next if line.is_a? HeadingLine
+
           vtt.cue(line.start_ms, line.end_ms, line.line)
         end
       end
 
       return vtt
     end
-
-    private def self.parse(initial_data : Hash(String, JSON::Any))
-      body = initial_data.dig("actions", 0, "updateEngagementPanelAction", "content", "transcriptRenderer",
-        "content", "transcriptSearchPanelRenderer", "body", "transcriptSegmentListRenderer",
-        "initialSegments").as_a
-
-      lines = [] of TranscriptLine
-      body.each do |line|
-        # Transcript section headers. They are not apart of the captions and as such we can safely skip them.
-        if line.as_h.has_key?("transcriptSectionHeaderRenderer")
-          next
-        end
-
-        line = line["transcriptSegmentRenderer"]
-
-        start_ms = line["startMs"].as_s.to_i.millisecond
-        end_ms = line["endMs"].as_s.to_i.millisecond
-
-        text = extract_text(line["snippet"]) || ""
-
-        lines << TranscriptLine.new(start_ms, end_ms, text)
-      end
-
-      return lines
-    end
   end
 end