Our project needs podcast episodes to have timecoded transcripts. Most podcasts don't provide that already. Let's use a transcription service to get rough transcripts. Although these automated solutions would not be acceptable as replacements for the audio, they tend to be good enough as guidelines to indicate which parts of a podcast episode contain the search term(s) that someone is using Steno.fm to look for.
Steno.fm needs to get the following for each podcast episode:
English text of recognized words from the audio stream, ideally with high accuracy and coverage of the entire episode.
Periodic time markers that relate time of the audio stream to position in the text.
Speaker attribution that relates the identity of the speaker to the text, ideally with every single word attributed and with all attributions correct.
List of keywords from the title and description of the podcast and the particular episode.
derek-schultz
commented
9 years ago
Our project needs podcast episodes to have timecoded transcripts. Most podcasts don't provide that already. Let's use a transcription service to get rough transcripts. Although these automated solutions would not be acceptable as replacements for the audio, they tend to be good enough as guidelines to indicate which parts of a podcast episode contain the search term(s) that someone is using Steno.fm to look for.
Steno.fm needs to get the following for each podcast episode:
Option: IBM Watson Speech to Text Service
Option: Kaldi