Hi,
I have a wav file of an audio book and I have its text.
I need to find location of each word in the audio file, that is where each word starts and ends, to split the wav file into many small wav files containing one word each. This is not a speech recognition task per se (because I know the text in advance) but it looks related.
How can I use SPHINX to do this?
Thanks a lot in advance.
Any help is greatly appreciated.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Thanks a lot Nickolay.
I am so glad that you have this program.
Now I have a second problem:
I have found a lot of web pages where they mention that they use sphinx3_align, forced-alignment or viterbi-alignment. But no mention of how to install or use it. Could you please give me a hint where to start?
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Thanks a lot again Nickolay,
Using your hints I have found sphinx3/src/tests/regression/test-align.sh/test-align.sh file.
And also sphinx3/model/lm/an4/pittsburgh.littleendian.raw which looks like an input audio file for the test-align.sh script.
The problem is that I can't find an output file that would contain word boundary information. Should I use some additional command line argument or change the script somehow to enable it?
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Hi,
I have a wav file of an audio book and I have its text.
I need to find location of each word in the audio file, that is where each word starts and ends, to split the wav file into many small wav files containing one word each. This is not a speech recognition task per se (because I know the text in advance) but it looks related.
How can I use SPHINX to do this?
Thanks a lot in advance.
Any help is greatly appreciated.
With sphinx3_align you can align text to transcripts and get word boundaries.
Thanks a lot Nickolay.
I am so glad that you have this program.
Now I have a second problem:
I have found a lot of web pages where they mention that they use sphinx3_align, forced-alignment or viterbi-alignment. But no mention of how to install or use it. Could you please give me a hint where to start?
> But no mention of how to install or use it.
It's in sphinx3 package
https://sourceforge.net/project/showfiles.php?group_id=1904&package_id=68406
you will also need acoustic model for your language and a dictionary
> Could you please give me a hint where to start?
Start with tests and demos from the package.
Thanks a lot again Nickolay,
Using your hints I have found sphinx3/src/tests/regression/test-align.sh/test-align.sh file.
And also sphinx3/model/lm/an4/pittsburgh.littleendian.raw which looks like an input audio file for the test-align.sh script.
The problem is that I can't find an output file that would contain word boundary information. Should I use some additional command line argument or change the script somehow to enable it?
It creates the file pittsburgh.littleendian.wdseg with word segments. Time is measured in frames (0.01 s).
Thanks a lot Nickolay.
That's exactly what I was looking for.