Hello, I have made a dictionary which has some lexically stressed phones, like AH0,AH1 etc.
Log file in the decode stage says the following for the words which have such phones:
ERROR: "dict.c", line 195: Line 47630: Phone 'EY1' is mising in the acoustic model; word 'STATE' ignored
ERROR: "dict.c", line 195: Line 47633: Phone 'EY1' is mising in the acoustic model; word 'STATED' ignored
ERROR: "dict.c", line 195: Line 47671: Phone 'AE1' is mising in the acoustic model; word 'STATUTE' ignored
ERROR: "dict.c", line 195: Line 47682: Phone 'EY1' is mising in the acoustic model; word 'STAYED' ignored
ERROR: "dict.c", line 195: Line 47697: Phone 'EH1' is mising in the acoustic model; word 'STEADY' ignored
ERROR: "dict.c", line 195: Line 47729: Phone 'IY1' is mising in the acoustic model; word 'STEEL' ignored
ERROR: "dict.c", line 195: Line 47760: Phone 'EH1' is mising in the acoustic model; word 'STEMS' ignored
ERROR: "dict.c", line 195: Line 47771: Phone 'EH1' is mising in the acoustic model; word 'STEP' ignored
ERROR: "dict.c", line 195: Line 47864: Phone 'IH1' is mising in the acoustic model; word 'STILL' ignored
ERROR: "dict.c", line 195: Line 47919: Phone 'AA1' is mising in the acoustic model; word 'STOCK' ignored
ERROR: "dict.c", line 195: Line 47935: Phone 'AA1' is mising in the acoustic model; word 'STOCKS' ignored
ERROR: "dict.c", line 195: Line 47994: Phone 'AO1' is mising in the acoustic model; word 'STORES' ignored
ERROR: "dict.c", line 195: Line 48203: Phone 'AO1' is mising in the acoustic model; word 'STRONG' ignored
ERROR: "dict.c", line 195: Line 48209: Phone 'AO1' is mising in the acoustic model; word 'STRONGHOLDS' ignored
ERROR: "dict.c", line 195: Line 48217: Phone 'AH1' is mising in the acoustic model; word 'STRUCK' ignored
ERROR: "dict.c", line 195: Line 48220: Phone 'AH1' is mising in the acoustic model; word 'STRUCTURE' ignored
ERROR: "dict.c", line 195: Line 48438: Phone 'AH0' is mising in the acoustic model; word 'SUBSCRIBERS' ignored
Do i need to add these phones in the .phone file for the words to be included while decoding? Or any other way to include these words. Kindly help me with this.
Thanks and Regards,
Nikhil Shirwandkar
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Thanks Nickolay for your prompt response. I will try to train the model with those phones.
I am trying to improve the accuracy of the model. Is there any other way to improve the accuracy, other than adding words in the dictionary with lexically stressed phones?
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Sure, as soon as you provide enough details. Who are you, what is the project you are working on, what have you done, what data do you use, current accuracies, etc.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I have just completed my Mtech, working on a Speech to Text project using CMU Sphinx.
I am using librispeech audio data of 250 hours for training, and PDA data set of 3 hours for testing. Currently i am getting:
SER%: 49.8 WER%: 20.2
I am trying to improve accuracy by adding some unrecognized or wrongly recognized words from PDA dataset to the model.
Am I on the right path? or is there any other way to improve the results?
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I am trying to build a speech to text utility which would transcribe call recordings. For now, using librispeech data to get acquainted with the procedure.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I am currently working to submit the project synopsis for further studies, I went through the CMU Sphinx's website and forums and have trained the acoustic model.
I dont have any call recordings data for now, so I am using librispeech data for understanding the process and setting up a demo.
I am hoping to get guidance from you on increasing the accuracy, or atleast getting a right path, I am not getting where i am going wrong.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Hello, I have made a dictionary which has some lexically stressed phones, like AH0,AH1 etc.
Log file in the decode stage says the following for the words which have such phones:
ERROR: "dict.c", line 195: Line 47630: Phone 'EY1' is mising in the acoustic model; word 'STATE' ignored
ERROR: "dict.c", line 195: Line 47633: Phone 'EY1' is mising in the acoustic model; word 'STATED' ignored
ERROR: "dict.c", line 195: Line 47671: Phone 'AE1' is mising in the acoustic model; word 'STATUTE' ignored
ERROR: "dict.c", line 195: Line 47682: Phone 'EY1' is mising in the acoustic model; word 'STAYED' ignored
ERROR: "dict.c", line 195: Line 47697: Phone 'EH1' is mising in the acoustic model; word 'STEADY' ignored
ERROR: "dict.c", line 195: Line 47729: Phone 'IY1' is mising in the acoustic model; word 'STEEL' ignored
ERROR: "dict.c", line 195: Line 47760: Phone 'EH1' is mising in the acoustic model; word 'STEMS' ignored
ERROR: "dict.c", line 195: Line 47771: Phone 'EH1' is mising in the acoustic model; word 'STEP' ignored
ERROR: "dict.c", line 195: Line 47864: Phone 'IH1' is mising in the acoustic model; word 'STILL' ignored
ERROR: "dict.c", line 195: Line 47919: Phone 'AA1' is mising in the acoustic model; word 'STOCK' ignored
ERROR: "dict.c", line 195: Line 47935: Phone 'AA1' is mising in the acoustic model; word 'STOCKS' ignored
ERROR: "dict.c", line 195: Line 47994: Phone 'AO1' is mising in the acoustic model; word 'STORES' ignored
ERROR: "dict.c", line 195: Line 48203: Phone 'AO1' is mising in the acoustic model; word 'STRONG' ignored
ERROR: "dict.c", line 195: Line 48209: Phone 'AO1' is mising in the acoustic model; word 'STRONGHOLDS' ignored
ERROR: "dict.c", line 195: Line 48217: Phone 'AH1' is mising in the acoustic model; word 'STRUCK' ignored
ERROR: "dict.c", line 195: Line 48220: Phone 'AH1' is mising in the acoustic model; word 'STRUCTURE' ignored
ERROR: "dict.c", line 195: Line 48438: Phone 'AH0' is mising in the acoustic model; word 'SUBSCRIBERS' ignored
Do i need to add these phones in the .phone file for the words to be included while decoding? Or any other way to include these words. Kindly help me with this.
Thanks and Regards,
Nikhil Shirwandkar
You have to train an acoustic model with those phones.
Thanks Nickolay for your prompt response. I will try to train the model with those phones.
I am trying to improve the accuracy of the model. Is there any other way to improve the accuracy, other than adding words in the dictionary with lexically stressed phones?
Yes
Can you please suggest one?
Sure, as soon as you provide enough details. Who are you, what is the project you are working on, what have you done, what data do you use, current accuracies, etc.
I have just completed my Mtech, working on a Speech to Text project using CMU Sphinx.
I am using librispeech audio data of 250 hours for training, and PDA data set of 3 hours for testing. Currently i am getting:
SER%: 49.8 WER%: 20.2
I am trying to improve accuracy by adding some unrecognized or wrongly recognized words from PDA dataset to the model.
Am I on the right path? or is there any other way to improve the results?
No
Could you elaborate on whats going wrong and guide me with this?
I am new to CMU Sphinx.
You didn't provide details about project you are working.
I am trying to build a speech to text utility which would transcribe call recordings. For now, using librispeech data to get acquainted with the procedure.
Then you are doing all wrong it seems.
I am currently working to submit the project synopsis for further studies, I went through the CMU Sphinx's website and forums and have trained the acoustic model.
I dont have any call recordings data for now, so I am using librispeech data for understanding the process and setting up a demo.
I am hoping to get guidance from you on increasing the accuracy, or atleast getting a right path, I am not getting where i am going wrong.