thavelick/youtranscript

"""
Get transcripts from youtube.

Portions of code here taken from:
* https://github.com/AKGV1/ytseach (MIT) and
* https://github.com/NoahCardoza/noter (unlicensed)

Once Invidious is updated to use the innertube api, for transcripts
much of the code here can be removed.
"""

import json
from collections import deque
from dataclasses import dataclass
from urllib.request import Request, urlopen
from pprint import pprint

YOUTUBE_URL = "https://www.youtube.com"


@dataclass
class TranscriptCue:
    """A transcript cue."""

    duration: float  # Length of the transcript cue in seconds.
    start: float  # Time into the video when the cue starts in seconds.
    text: str  # Spoken text

    def start_time_text(self) -> str:
        """Return the start time of the cue in the form minutes:ss."""
        start = int(self.start)
        minute_part = start // 60
        second_part = start % 60
        return f'{minute_part}:{second_part:02d}'

    def youtube_link_tag(self, youtube_id: str) -> str:
        """Return a html link to the youtube video at the cue start time."""
        return (
            f'<a href="{YOUTUBE_URL}/watch'
            f'?v={youtube_id}&t={self.start}">{self.start_time_text()}</a>'
        )

    def html_text(self):
        """Return the text of the cue as html."""
        return self.text.replace('\n', '<br>')


def _consolidate_cues(
        cues: list[TranscriptCue],
        max_duration: int) -> list[TranscriptCue]:
    """Consolidate cues into a bigger chunks of text."""
    if len(cues) == 0:
        return []

    consolidated_cues: list[TranscriptCue] = []
    cues_remaining = deque(cues)
    bigger_cue = TranscriptCue(
        duration=0,
        start=cues[0].start,
        text=''
    )
    total_duration = 0.0

    while len(cues_remaining) > 0:
        cue = cues_remaining.popleft()
        if bigger_cue.duration + cue.duration > max_duration:
            consolidated_cues.append(bigger_cue)
            bigger_cue = TranscriptCue(
                duration=0,
                start=cue.start,
                text='',
            )

        total_duration += cue.duration
        bigger_cue.duration += cue.duration
        bigger_cue.text += f"{cue.text}\n"

    consolidated_cues.append(bigger_cue)
    return consolidated_cues


def _get_youtube_page(url: str) -> str:
    """Get the page for a youtube url."""
    video_page_request = Request(url)
    with urlopen(video_page_request) as response:
        text = response.read().decode('utf-8')
        if "This video is'nt available anymore" in text:
            raise ValueError("video-page-not-found")
        elif response.status != 200:
            raise ValueError("http-error-fetching-video-page")
    return text


def _parse_value_from_page_text(text: str, key: str) -> str:
    """Get the value for a key from a page text."""
    return text.split(f'"{key}":"')[1].split('"')[0]


def _get_transcript_from_innertube_api(
        serialized_share_entity: str,
        inner_tube_api_key: str) -> dict:
    """
    Get a transcript from the innertube api.

    Args:
        serialized_share_entity: (whatever that is).
            It is just a string of random encoded data parsed from the youtube
            page.
        inner_tube_api_key: The innertube api key. A random string of
            characters that is parsed out of the youtube page.
    """
    user_agent = (
        # Probably doesn't matter that much. Just make it look like a browser.
        'Mozilla/5.0 (Windows NT 10.0; Win64; x64) '
        'AppleWebKit/537.36 (KHTML, like Gecko) '
        'Chrome/97.0.4692.99 Safari/537.36 OPR/83.0.4254.70,gzip(gfe)'
    )
    payload = {
        "context": {
            "client": {
                "userAgent": user_agent,
                "clientName": "WEB",
                "clientVersion": "2.20220309.01.00",
            }
        },
        "params": serialized_share_entity
    }

    transcript_request = Request(
        f'{YOUTUBE_URL}/youtubei/v1/get_transcript?key={inner_tube_api_key}',
        json.dumps(payload).encode('utf-8'),
        {'Content-Type': 'application/json'},
    )

    with urlopen(transcript_request) as response:
        if response.status != 200:
            raise ValueError("http-error-fetching-transcript")
        transcript_data = json.loads(response.read())
    return transcript_data


def get_transcript(youtube_id: str) -> list[TranscriptCue]:
    """
    Get a transcript for a youtube url using the innertube api.

    Args:
        youtube_id: The youtube id of the video.
    Returns:
        A list of TranscriptCues.
    """
    transcript_log = []

    text = _get_youtube_page(f"{YOUTUBE_URL}/watch?v={youtube_id}")

    innertube_api_key = _parse_value_from_page_text(text, 'INNERTUBE_API_KEY')
    serialized_share_entity = _parse_value_from_page_text(
        text,
        'serializedShareEntity'
    )
    transcript_data = _get_transcript_from_innertube_api(
        serialized_share_entity,
        innertube_api_key
    )

    if not transcript_data.get('actions'):
        return []

    for action in transcript_data['actions']:
        for group in (
                action['updateEngagementPanelAction']['content']
                ['transcriptRenderer']['body']['transcriptBodyRenderer']
                ['cueGroups']):
            for cue in group['transcriptCueGroupRenderer']['cues']:
                renderer = cue['transcriptCueRenderer']
                duration = float(renderer['durationMs']) / 1000
                start_time = float(renderer['startOffsetMs']) / 1000
                text = renderer['cue']['simpleText']
                transcript_log.append(
                    TranscriptCue(duration, start_time, text)
                )

    return _consolidate_cues(transcript_log, max_duration=30)


if __name__ == '__main__':
    pprint(
        get_transcript('yg9HwXI6Ha0')
    )