Transcribe podcasts to make them searchable, and then extract audio clips based on search terms.
I initially wrote this to make it easier to make drops for the Doughboys podcast.
To download and transcribe an entire podcast:
php fetch.php --feed http://example.com/feed.xml
To find all occurrences of someone saying "have a great day", extract the audio plus an additional second on each side, and save the clips to the directory great-day/,
php search.php --search "have a great day" --extract --before 1 --after 1 --output_dir great-day/
The transcription portion requires the audio transcription tool whisper. It is easy to install and free to use.
To extract clips, you'll need ffmpeg. It is also easy to install and free to use.
You can also use the whisper.cpp port of whisper. It is free to use, but not as easy to install. It is, however, much faster in certain environments than the original whisper command line tool.
If you have whisper.cpp installed, you can tell Dropseeker to use it by specifying the --whisper_cpp /path/to/whisper.cpp/directory/ command line option.
For fetch.php:
Required:
--feed [url] The URL of a podcast RSS feed.
Optional:
--after_date [YYYY-MM-DD] Only download/transcribe episodes published after this date.
--before_date [YYYY-MM-DD] Only download/transcribe episodes published before this date.
--confirm Require confirmation before downloading or transcribing an episode.
--episode_dir [path] The directory in which to store the episode directories.
--exclude [string] Don't download episodes that match this string. If multiple --exclude strings are supplied, it will exclude any episodes that match any of the supplied strings.
--fetch_only Just download episodes; don't transcribe.
--help Show the usage instructions.
--include [string] Only download/transcribe episodes that match this string. If multiple --include strings are supplied, it will include any episodes that match any of the supplied strings.
--title [string] The string that should be used for the folders containing the recordings and transcripts.
--transcript_dir [path] The directory in which to store the transcript directories.
--transcribe_only Just transcribe episodes; don't download any new ones.
--whisper_cpp [path] The path to whisper.cpp's installation folder, if you want to use it instead of the standard whisper tool. This folder should contain the `main` executable.
--whisper_[arg] [?arg] Pass the command line option `arg` to the `whisper` executable. e.g., `--whisper_model medium` will call run `whisper --model medium`.
Default whisper args:
--model tiny
--output_dir The subdirectory of --transcript_dir with the same name as --title.
For search.php
Required:
--search [string] What to search for in transcripts. Supports wildcards like 'foo*' (words that start
with foo), 'foo * bar' ('foo' and 'bar' separated by one word), or 'foo*baz*bar' (any
word starting with 'foo', containing 'baz', and ending with 'bar').
Optional:
--after [float] Extract an additional __ seconds from after each match.
--before [float] Extract an additional __ seconds from before each match.
--episode_dir [path] The directory in which the episode directories are stored, if not in the default location.
--extract Extract audio clips of each match.
--help Show the usage instructions.
--context [string] Only consider a match if the full prefix + match + suffix also includes this string.
--context_exclude [string] A search string that, if it matches text around the search result, will be excluded from the final results.
--output_dir [path] The directory in which to store the extracted audio clips.
--limit [int] Stop searching entirely after finding this many total matches.
--limit_per_episode [int] Stop searching an episode after finding this many matches in it.
--match [string] Only check episodes that include this string in their filename.
--min_duration [float] If extracting audio, only extract a clip if it will be at least this long.
--podcast [string] Only search transcripts from podcasts that include this string in their title.
--prefix_words [int] Show this many words before the matching string in the text search results.
--suffix_words [int] Show this many words after the matching string in the text search results.
--transcript_dir [path] The directory in which the transcript directories are stored, if not in the default location.
If you want to specify a default set of command line options different from what Dropseeker specifies, you can do so by creating a file called dropseeker.conf in this directory. Add a line like this for each option you want to specify as a default:
key=value
or for options that don't take a value, just
key
For example, this is a valid dropseeker.conf file:
episode_dir=/path/to/episode/dir/
before=2
after=5
confirm
You can still specify different values for these options on the command line that will overwrite the values you listed in dropseeker.conf.