SimpleAudioIndexer 1.0.0

Command-line Usage

Open up a terminal and enter.

$ sai --mode "ibm" --username_ibm USERNAME --password_ibm PASSWORD --src_dir SRC_DIR --search "called"

{'called': {'small_audio.wav': [(1.25, 1.71)]}}

Replace USERNAME and PASSWORD with your IBM Watson’s credentials and SRC_DIR with the absolute path to the directory you just prepared.

The out would be, like above, a dictionary that has the query, the file(s) it appears in and the all of the (starting second, ending second) of that query.

Note that all commands work uniformally for other engines (i.e. Pocketsphinx), for example the command above can be enterred as:

$ sai --mode "cmu" --src_dir SRC_DIR --search "lives"

{'our': {'small_audio': [(2.93, 3.09)]}}

Which would use Pocketsphinx instead of Watson to get the timestamps. Note that the quality/accuracy of Pocketsphinx is much lower than Watson.

Instead of searching for a word, you could also match a regex pattern, for example:

$ sai --mode ibm --src_dir SRC_DIR --username_ibm USERNAME --password_ibm PASSWORD --regexp " [a-z][a-z] "

{u' in ': {'small_audio.wav': [(2.81, 2.93)]},
{u' to ': {'small_audio.wav': [(1.71, 1.81)]}}

That was the result of searching for two letter words. Note that your results would match any aribtrary regular expressions.

You may also save and load the indexed data from the command line script. For more information, visit here.

Library Usage

Say you have this file

>>> from SimpleAudioIndexer import SimpleAudioIndexer as sai

Afterwards, you should create an instance of sai

>>> indexer = sai(mode="ibm", src_dir="SRC_DIR", username_ibm="USERNAME", password_ibm="PASSWORD")

Now you may index all the available audio files by calling index_audio method:

>>> indexer.index_audio()

You could have a searching generator:

>>> searcher = indexer.search_gen(query="called")
# If you're on python 2.7, instead of below, do print searcher.next()
>>> print(next(searcher))
{'Query': 'called', 'File Name': 'small_audio.wav', 'Result': (1.25, 1.71)}

Now there are quite a few more arguments implemented for search_gen. Say you wanted your search to be case sensitive (by default it’s not). Or, say you wanted to look for a phrase but there’s a timing gap and the indexer didn’t pick it up right, you could specify timing_error. Or, say some word is completely missed, then you could specify missing_word_tolerance etc.

For a full list, see the API reference here

Note that you could also call search_all method to have search for a list of queries within all the audio files:

Finally, you could do a regex search!

>>> print(indexer.search_regexp(pattern="[A-Z][^l]* ")
{u'Americans are ca': {'small_audio.wav': [(0.21, 1.71)]}}

There are more functionalities implemented. For detailed explainations, read the documentation here.