0
0
Fork 0
mirror of https://github.com/yt-dlp/yt-dlp.git synced 2024-11-18 01:35:12 +00:00
yt-dlp/yt_dlp/extractor/sverigesradio.py
Pccode66 7a5c1cfe93
Completely change project name to yt-dlp (#85)
* All modules and binary names are changed
* All documentation references changed
* yt-dlp no longer loads youtube-dlc config files
* All URLs changed to point to organization account

Co-authored-by: Pccode66
Co-authored-by: pukkandan
2021-02-25 00:15:56 +05:30

115 lines
3.8 KiB
Python

# coding: utf-8
from __future__ import unicode_literals
from .common import InfoExtractor
from ..utils import (
determine_ext,
int_or_none,
str_or_none,
)
class SverigesRadioBaseIE(InfoExtractor):
_BASE_URL = 'https://sverigesradio.se/sida/playerajax/'
_QUALITIES = ['low', 'medium', 'high']
_EXT_TO_CODEC_MAP = {
'mp3': 'mp3',
'm4a': 'aac',
}
_CODING_FORMAT_TO_ABR_MAP = {
5: 128,
11: 192,
12: 32,
13: 96,
}
def _real_extract(self, url):
audio_id = self._match_id(url)
query = {
'id': audio_id,
'type': self._AUDIO_TYPE,
}
item = self._download_json(
self._BASE_URL + 'audiometadata', audio_id,
'Downloading audio JSON metadata', query=query)['items'][0]
title = item['subtitle']
query['format'] = 'iis'
urls = []
formats = []
for quality in self._QUALITIES:
query['quality'] = quality
audio_url_data = self._download_json(
self._BASE_URL + 'getaudiourl', audio_id,
'Downloading %s format JSON metadata' % quality,
fatal=False, query=query) or {}
audio_url = audio_url_data.get('audioUrl')
if not audio_url or audio_url in urls:
continue
urls.append(audio_url)
ext = determine_ext(audio_url)
coding_format = audio_url_data.get('codingFormat')
abr = int_or_none(self._search_regex(
r'_a(\d+)\.m4a', audio_url, 'audio bitrate',
default=None)) or self._CODING_FORMAT_TO_ABR_MAP.get(coding_format)
formats.append({
'abr': abr,
'acodec': self._EXT_TO_CODEC_MAP.get(ext),
'ext': ext,
'format_id': str_or_none(coding_format),
'vcodec': 'none',
'url': audio_url,
})
self._sort_formats(formats)
return {
'id': audio_id,
'title': title,
'formats': formats,
'series': item.get('title'),
'duration': int_or_none(item.get('duration')),
'thumbnail': item.get('displayimageurl'),
'description': item.get('description'),
}
class SverigesRadioPublicationIE(SverigesRadioBaseIE):
IE_NAME = 'sverigesradio:publication'
_VALID_URL = r'https?://(?:www\.)?sverigesradio\.se/sida/(?:artikel|gruppsida)\.aspx\?.*?\bartikel=(?P<id>[0-9]+)'
_TESTS = [{
'url': 'https://sverigesradio.se/sida/artikel.aspx?programid=83&artikel=7038546',
'md5': '6a4917e1923fccb080e5a206a5afa542',
'info_dict': {
'id': '7038546',
'ext': 'm4a',
'duration': 132,
'series': 'Nyheter (Ekot)',
'title': 'Esa Teittinen: Sanningen har inte kommit fram',
'description': 'md5:daf7ce66a8f0a53d5465a5984d3839df',
'thumbnail': r're:^https?://.*\.jpg',
},
}, {
'url': 'https://sverigesradio.se/sida/gruppsida.aspx?programid=3304&grupp=6247&artikel=7146887',
'only_matching': True,
}]
_AUDIO_TYPE = 'publication'
class SverigesRadioEpisodeIE(SverigesRadioBaseIE):
IE_NAME = 'sverigesradio:episode'
_VALID_URL = r'https?://(?:www\.)?sverigesradio\.se/(?:sida/)?avsnitt/(?P<id>[0-9]+)'
_TEST = {
'url': 'https://sverigesradio.se/avsnitt/1140922?programid=1300',
'md5': '20dc4d8db24228f846be390b0c59a07c',
'info_dict': {
'id': '1140922',
'ext': 'mp3',
'duration': 3307,
'series': 'Konflikt',
'title': 'Metoo och valen',
'description': 'md5:fcb5c1f667f00badcc702b196f10a27e',
'thumbnail': r're:^https?://.*\.jpg',
}
}
_AUDIO_TYPE = 'episode'