search

santi-pdp / pase

Problem Agnostic Speech Encoder
MIT License
439 stars 87 forks source link

ASR experiment on TIMIT dataset #82

Open zengchang233 opened 4 years ago

zengchang233 commented 4 years ago

Hi, thanks for your nice work.

I am trying ASR experiment on TIMIT dataset, but I got an error message as the below.

epoch=20 loss_tr=0.859986 err_tr=0.273605 loss_te=1.420085 err_te=0.400905 lr=0.000562
epoch=21 loss_tr=0.856079 err_tr=0.271644 loss_te=1.421840 err_te=0.401584 lr=0.000281
epoch=22 loss_tr=0.855693 err_tr=0.271957 loss_te=1.419042 err_te=0.400856 lr=0.000141
epoch=23 loss_tr=0.854678 err_tr=0.271677 loss_te=1.419402 err_te=0.400407 lr=0.000070
BEST ERR=0.400407
BEST ACC=0.599593
WARNING: QRNN ignores bidirectional flag
Waveform reading...
Computing PASE features...
Decoding...
kaldi_decoding_scripts//decode_dnn.sh /home/sysadmin/ncr/pase/pase-test/ASR/output/dec_cfg.ini ./output/dec "./output/post.ark"
Traceback (most recent call last):
  File "run_TIMIT_full_decoding.py", line 559, in <module>
    run_shell(cmd_decode)
  File "/home/sysadmin/ncr/pase/pase-test/ASR/utils.py", line 50, in run_shell
    p = subprocess.Popen(cmd, stdout=subprocess.PIPE, stderr=subprocess.PIPE,shell=True)
  File "/home/sysadmin/miniconda/envs/pase/lib/python3.7/subprocess.py", line 800, in __init__
    restore_signals, start_new_session)
  File "/home/sysadmin/miniconda/envs/pase/lib/python3.7/subprocess.py", line 1482, in _execute_child
    restore_signals, start_new_session, preexec_fn)
OSError: [Errno 12] Cannot allocate memory

Do you have any idea about this?

Thanks so much.

mravanelli commented 4 years ago

Hi, the current version of the ASR script is memory demanding because it stores the waveforms in RAM. However, the decoding part shouldn't be very memory demanding. Could you please try to run the script with decoding_only = False?

Mirco

On Mon, 10 Feb 2020 at 05:42, zengchang94622 notifications@github.com wrote:

Hi, thanks for your nice work.

I am trying ASR experiment on TIMIT dataset, but I got an error message as the below.

epoch=20 loss_tr=0.859986 err_tr=0.273605 loss_te=1.420085 err_te=0.400905 lr=0.000562 epoch=21 loss_tr=0.856079 err_tr=0.271644 loss_te=1.421840 err_te=0.401584 lr=0.000281 epoch=22 loss_tr=0.855693 err_tr=0.271957 loss_te=1.419042 err_te=0.400856 lr=0.000141 epoch=23 loss_tr=0.854678 err_tr=0.271677 loss_te=1.419402 err_te=0.400407 lr=0.000070 BEST ERR=0.400407 BEST ACC=0.599593 WARNING: QRNN ignores bidirectional flag Waveform reading... Computing PASE features... Decoding... kaldi_decoding_scripts//decode_dnn.sh /home/sysadmin/ncr/pase/pase-test/ASR/output/dec_cfg.ini ./output/dec "./output/post.ark" Traceback (most recent call last): File "run_TIMIT_full_decoding.py", line 559, in run_shell(cmd_decode) File "/home/sysadmin/ncr/pase/pase-test/ASR/utils.py", line 50, in run_shell p = subprocess.Popen(cmd, stdout=subprocess.PIPE, stderr=subprocess.PIPE,shell=True) File "/home/sysadmin/miniconda/envs/pase/lib/python3.7/subprocess.py", line 800, in init restore_signals, start_new_session) File "/home/sysadmin/miniconda/envs/pase/lib/python3.7/subprocess.py", line 1482, in _execute_child restore_signals, start_new_session, preexec_fn) OSError: [Errno 12] Cannot allocate memory

Do you have any idea about this?

Thanks so much.

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/santi-pdp/pase/issues/82?email_source=notifications&email_token=AEA2ZVTCHXXMMRF4LFCDIBLRCEVQTA5CNFSM4KSLV4O2YY3PNVWWK3TUL52HS4DFUVEXG43VMWVGG33NNVSW45C7NFSM4IMGVLYQ, or unsubscribe https://github.com/notifications/unsubscribe-auth/AEA2ZVSOQVMSNO2E2ZZUF6TRCEVQTANCNFSM4KSLV4OQ .

mravanelli commented 4 years ago

Sorry, with decoding_only = True (line 421)

On Mon, 10 Feb 2020 at 09:59, Mirco Ravanelli mirco.ravanelli@gmail.com wrote:

Hi, the current version of the ASR script is memory demanding because it stores the waveforms in RAM. However, the decoding part shouldn't be very memory demanding. Could you please try to run the script with decoding_only = False?

Mirco

On Mon, 10 Feb 2020 at 05:42, zengchang94622 notifications@github.com wrote:

Hi, thanks for your nice work.

I am trying ASR experiment on TIMIT dataset, but I got an error message as the below.

epoch=20 loss_tr=0.859986 err_tr=0.273605 loss_te=1.420085 err_te=0.400905 lr=0.000562 epoch=21 loss_tr=0.856079 err_tr=0.271644 loss_te=1.421840 err_te=0.401584 lr=0.000281 epoch=22 loss_tr=0.855693 err_tr=0.271957 loss_te=1.419042 err_te=0.400856 lr=0.000141 epoch=23 loss_tr=0.854678 err_tr=0.271677 loss_te=1.419402 err_te=0.400407 lr=0.000070 BEST ERR=0.400407 BEST ACC=0.599593 WARNING: QRNN ignores bidirectional flag Waveform reading... Computing PASE features... Decoding... kaldi_decoding_scripts//decode_dnn.sh /home/sysadmin/ncr/pase/pase-test/ASR/output/dec_cfg.ini ./output/dec "./output/post.ark" Traceback (most recent call last): File "run_TIMIT_full_decoding.py", line 559, in run_shell(cmd_decode) File "/home/sysadmin/ncr/pase/pase-test/ASR/utils.py", line 50, in run_shell p = subprocess.Popen(cmd, stdout=subprocess.PIPE, stderr=subprocess.PIPE,shell=True) File "/home/sysadmin/miniconda/envs/pase/lib/python3.7/subprocess.py", line 800, in init restore_signals, start_new_session) File "/home/sysadmin/miniconda/envs/pase/lib/python3.7/subprocess.py", line 1482, in _execute_child restore_signals, start_new_session, preexec_fn) OSError: [Errno 12] Cannot allocate memory

Do you have any idea about this?

Thanks so much.

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/santi-pdp/pase/issues/82?email_source=notifications&email_token=AEA2ZVTCHXXMMRF4LFCDIBLRCEVQTA5CNFSM4KSLV4O2YY3PNVWWK3TUL52HS4DFUVEXG43VMWVGG33NNVSW45C7NFSM4IMGVLYQ, or unsubscribe https://github.com/notifications/unsubscribe-auth/AEA2ZVSOQVMSNO2E2ZZUF6TRCEVQTANCNFSM4KSLV4OQ .

zengchang233 commented 4 years ago

Thanks for reply. It works!

But there is another problem. 

Decoding...
kaldi_decoding_scripts/decode_dnn.sh /home/sysadmin/ncr/pase/pase-test/ASR/output/dec_cfg.ini ./output/dec "./output/post.ark"
Decoded

After decoding, I think it will generate the result in ./output/dec directory according to this command and RESULTS script. But it doesn't.

Could you give me some advice to fix it? Thanks

mravanelli commented 4 years ago

Could you replace the output folder with its absolute path? We noticed some problems with relative paths when calling the decoder.

Best,

Mirco

On Mon, 10 Feb 2020 at 21:41, zengchang94622 notifications@github.com wrote:

Thanks for reply. It works!

But there is another problem.

kaldi_decoding_scripts/decode_dnn.sh /home/sysadmin/ncr/pase/pase-test/ASR/output/dec_cfg.ini ./output/dec "./output/post.ark" Decoded

After decoding, I think it will generate the result in ./output/dec directory according to this command and RESULTS script. But it doesn't.

Could you give me some advice to fix it? Thanks

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/santi-pdp/pase/issues/82?email_source=notifications&email_token=AEA2ZVRBZ62PFPIHZ7JTODDRCIF4NA5CNFSM4KSLV4O2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOELLBUAA#issuecomment-584456704, or unsubscribe https://github.com/notifications/unsubscribe-auth/AEA2ZVQZ2DMIXTBFWI7N2HDRCIF4NANCNFSM4KSLV4OQ .

mravanelli commented 4 years ago

I did some changes. Could you try it with the latest code and let us know?

Thank you!

Best,

Mirco

On Mon, 10 Feb 2020 at 21:55, Mirco Ravanelli mirco.ravanelli@gmail.com wrote:

Could you replace the output folder with its absolute path? We noticed some problems with relative paths when calling the decoder.

Best,

Mirco

On Mon, 10 Feb 2020 at 21:41, zengchang94622 notifications@github.com wrote:

Thanks for reply. It works!

But there is another problem.

kaldi_decoding_scripts/decode_dnn.sh /home/sysadmin/ncr/pase/pase-test/ASR/output/dec_cfg.ini ./output/dec "./output/post.ark" Decoded

After decoding, I think it will generate the result in ./output/dec directory according to this command and RESULTS script. But it doesn't.

Could you give me some advice to fix it? Thanks

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/santi-pdp/pase/issues/82?email_source=notifications&email_token=AEA2ZVRBZ62PFPIHZ7JTODDRCIF4NA5CNFSM4KSLV4O2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOELLBUAA#issuecomment-584456704, or unsubscribe https://github.com/notifications/unsubscribe-auth/AEA2ZVQZ2DMIXTBFWI7N2HDRCIF4NANCNFSM4KSLV4OQ .

zengchang233 commented 4 years ago

I did some changes. Could you try it with the latest code and let us know? Thank you! Best, Mirco On Mon, 10 Feb 2020 at 21:55, Mirco Ravanelli mirco.ravanelli@gmail.com wrote: … Could you replace the output folder with its absolute path? We noticed some problems with relative paths when calling the decoder. Best, Mirco On Mon, 10 Feb 2020 at 21:41, zengchang94622 @.***> wrote: > Thanks for reply. It works! > > But there is another problem. > > kaldi_decoding_scripts/decode_dnn.sh /home/sysadmin/ncr/pase/pase-test/ASR/output/dec_cfg.ini ./output/dec "./output/post.ark" > Decoded > > After decoding, I think it will generate the result in ./output/dec > directory according to this command and RESULTS script. But it doesn't. > > Could you give me some advice to fix it? > Thanks > > — > You are receiving this because you commented. > Reply to this email directly, view it on GitHub > <#82?email_source=notifications&email_token=AEA2ZVRBZ62PFPIHZ7JTODDRCIF4NA5CNFSM4KSLV4O2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOELLBUAA#issuecomment-584456704>, > or unsubscribe > https://github.com/notifications/unsubscribe-auth/AEA2ZVQZ2DMIXTBFWI7N2HDRCIF4NANCNFSM4KSLV4OQ > . >

Thanks. I will try the new one!

008karan commented 4 years ago

hows the ASR performance @zengchang94622