Scribd
Upload
71 views2 pages

Flatstart - Training - Exe From The Respective .C Files

The document provides instructions for training and decoding models using HTK. It outlines 10 steps for training, including installing HTK, generating a word network from a grammar file, creating MFCC features from audio files, and running the flatstart_training executable. It also outlines 3 steps for decoding, including ensuring a broad class mapping file is present, generating a word network, and running the Recognition_Decode_Script with proper paths and filenames.

Uploaded by

hkpn
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOC, PDF, TXT or read online on Scribd
71 views2 pages

Flatstart - Training - Exe From The Respective .C Files

The document provides instructions for training and decoding models using HTK. It outlines 10 steps for training, including installing HTK, generating a word network from a grammar file, creating MFCC features from audio files, and running the flatstart_training executable. It also outlines 3 steps for decoding, including ensuring a broad class mapping file is present, generating a word network, and running the Recognition_Decode_Script with proper paths and filenames.

Uploaded by

hkpn
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOC, PDF, TXT or read online on Scribd
You are on page 1/ 2

HTK

Training 1) Make sure that HTK is properly installed by typing the command HInit at the prompt 2) Make sure, if you are using broad class training, that the led file which shows the broad class mapping is created by the user and is present in the folder as mentioned in the flatstart_training.c file. 3) Generate the word network file from the grammar file using the HParse command

HParse <grammar_file_name > <word_network_file> 4) Generate the MFCC of all wave files which are supposed to be trained and decoded
HCopy s 100 e -100 -C <Configuration_file_name> -S <mfcc_file_name> The mfcc_file_name should have the format shown below F:\HKP\HTK\Data\s27g2713.wav F:\HKP\HTK\Data\s27g2713.mfc F:\HKP\HTK\Data\s27g2714.wav F:\HKP\HTK\Data\s27g2714.mfc F:\HKP\HTK\Data\s27g2715.wav F:\HKP\HTK\Data\s27g2715.mfc .........................................................

5) We will be giving these files as the input to the HTK 6) Create a file with the paths of all the MFCC files corresponding to the wave files 7) Create a file with the paths of all .phn files corresponding to the wave files 8) Generate the proto.exe, hmmdefs.exe , mixscript.exe and flatstart_training.exe from the respective .c files 9) Make sure that the paths and filenames in the above .c files are proper 10)Run the flatstart_training.exe now

Decoding 1) Make sure, if you are using broad class training, that the led file which shows the broad class mapping is created by the user and is present in the folder as mentioned in the Recognition_Decode_Script.m

2) Generate the word network file from the grammar file using the HParse command

HParse <grammar_file_name > <word_network_file> 3) Run the commands in Recognition_Decode_Script.m after making sure that the paths and filenames are proper

You might also like