pocketsphinx noise cancellation and best method to build an acoustic model?

[ab1984]

1. Is their any way to neglect any background noises like breathing or cough in pocketsphinx?
2. If their is any method, how can I do the noise cancellation?
3. Can I do it using .filler file?
4. So far I'm building the acoustic model by recording sentences and phrases by pronuncing them word by word and save the audio file for each and every sentence. Is that the better way or should I have to record them word by word and save the audio files seperatly?

Thank You

[bic-user]

1. yes. you can recognize them and discard
2. yes. Stationary noise cancellation is done durion feature extraction step.
3. .filler file contains fillers (breath, cough) that are included into your acoustic model
4. You can record whole thing and than separate audio automatically. But it's easier to record corpus sentence by sentence. You can fix audio or transcription once something goes wrong. To use fillers in your acoustic model you should add them into transcription if they occur in audio.

[ab1984]

Thank you for the reply bic-user. Still I'm having some confusions,
1. How should I do it?
3. How can I add new fillers to my acoustic model. (For an example Bus Horn Sound)
4. Is their a specific speed of word utterance when recording? Should I have to keep small breaks between two words or just record continuously as we are speaking?

Thank You

[bic-user]

1. In cmusphinx decoders it is done automatically. You can play with -fillprob parameter to control fillers insertion intensivity.
2. You need to retrain acoustic model adding training utterances with this filler
3. Just record in the same way, that is expected during recognition.

[ab1984]

Thank you bic-user.
Still no idea how to retrain the acoustic model to recognize such noise utterances. What are steps should I follow? Is their any tutorials to follow?

thank you