fgnt / pb_chime5 Goto Github PK

Hi,
I was trying to generate CHiME6 eval files using scripts from Kaldi and noticed that my script gets stuck at this stage .

python -m pb_chime5.database.chime5.create_json -j cache/chime6.json -db kaldi/egs/chime6/s5_track1/CHiME6 --transcription-path kaldi/kaldi/egs/chime6/s5_track1/CHiME6/transcriptions --chime6

which is called from here https://github.com/kaldi-asr/kaldi/blob/master/egs/chime6/s5b_track1/local/decode.sh#L102

Do you have any estimate of how long is the make command supposed to take. And, if there are any ways to speed up the process.

For example, simply using more threads for make command using make -j80 cache/chime6.json doesn't seem to help...

Thanks in advance !

Hello,
When i run activity_alignment.py(Generate finetuned time annotations from kaldi worn alignments), there is something wrong ，The following information is an error message：

Warning: Could not find P08_S02-0883960-0884024 in non_sil_alignment.
Warning: Could not find P08_S02-0887758-0888024 in non_sil_alignment.
Warning: Could not find P08_S02-0887758-0888024 in non_sil_alignment.
Warning: Could not find P08_S02-0887758-0888024 in non_sil_alignment.
Warning: Could not find P08_S02-0887758-0888024 in non_sil_alignment.
Warning: Could not find P08_S02-0887758-0888024 in non_sil_alignment.
Warning: Could not find P08_S02-0887758-0888024 in non_sil_alignment.
Warning: Could not find P08_S02-0889164-0889316 in non_sil_alignment.
Warning: Could not find P08_S02-0889164-0889316 in non_sil_alignment.
Warning: Could not find P08_S02-0889164-0889316 in non_sil_alignment.
Warning: Could not find P08_S02-0889164-0889316 in non_sil_alignment.
Warning: Could not find P08_S02-0889164-0889316 in non_sil_alignment.
Warning: Could not find P08_S02-0889164-0889316 in non_sil_alignment.
Traceback (most recent call last):
File "activity_alignment.py", line 198, in
use_ArrayIntervall=True,
File "/kaldi/egs/chime6/pb_chime5/pb_chime5/activity.py", line 214, in get_activity
f'Something went wrong.\n'
RuntimeError: Something went wrong.
Expected 22932 times a finetuned annotation for session S02, but 22932 times they are missing.
Expect that at least 1911 finetuned annotations are available, when non_sil_alignment_fn is given.
Otherwise assume something went wrong.

Thanks a lot if you can answer how to avoid this mistake

This part should be dlp_mpi (pointed by @yubouf)

hi,

Now, we're working on the CHiME-6 baseline, and try to include the GSS in the baseline recipe.
In the CHiME-6, we plan to synchronize the audio files across arrays and use the same time across them.
The audio files are modified accordingly and the JSON format will be changed as follows:

• CHiME-5

        "end_time": {
            "original": "0:01:00.38",
            "U01": "0:01:00.38",
            "U02": "0:01:00.39",
            "U03": "0:01:00.38",
            "U04": "0:01:00.37",
            "U05": "0:01:00.38",
            "U06": "0:01:00.36",
            "P09": "0:01:00.38",
            "P10": "0:01:00.38",
            "P11": "0:01:00.38",
            "P12": "0:01:00.38"
        },
        "start_time": {
            "original": "0:00:57.54",
            "U01": "0:00:57.54",
            "U02": "0:00:57.55",
            "U03": "0:00:57.54",
            "U04": "0:00:57.53",
            "U05": "0:00:57.54",
            "U06": "0:00:57.52",
            "P09": "0:00:57.54",
            "P10": "0:00:57.54",
            "P11": "0:00:57.54",
            "P12": "0:00:57.54"
        },
        "words": "[noise] What were we talking about again? [inaudible 0:00:58.96]",
        "speaker": "P12",
        "session_id": "S03"

• CHiME-6 (planning)

    {
        "end_time": "0:01:00.39",
        "start_time": "0:00:57.55",
        "words": "[noise] What were we talking about again? [inaudible 0:00:58.96]",
        "speaker": "P12",
        "session_id": "S03"
    },

Could you make an option to follow such a format?

In line 198 of file core_chime6.py, method predict is used with a parameter source_activity_mask

affiliation = cur.predict(
                   Obs.T[f, ...],
                   source_activity_mask=source_active_mask[f, ..., :T]
               )

But in the definition of the object cur (as well as class CACGMM in pb_bss/distribution/cacgmm.py), the method predict doesn't have this parameter. Simply changing predict to _predictdoesn't help.

Thanks a lot if you can answer when you are free

Hi, I am a newer in chime5 and I try to run your code recently. When I set the session_id = train , it would have key error about 'reference_array'. So I want to know if there is another parameter in this run command? Thank you ,and my run command is "mpiexec -np 9 python -m pb_chime5.scripts.run with session_id=train".

When eval transcripts are missing and EVAL_TRANSCRIPTIONS_MISSING is False (by default), there is no error message. However, an assertion is raised at this line.