TMDX5535EZDSP: TI Embedded Speech Recogniser training set generation

Abhinandan BR

Part Number: TMDX5535EZDSP

Hi,

I am working of TI ESR grammar creation and training models for TMDX5535EZDSP.

Command used:

mkdir Data/HMM12
HERest -C UserDelta.cnf -H Data/HMM11/triphones_fm.mmf \
-I Data/Triphones_f.mlf -I Data/Triphones_m.mlf \
-S Data/HTK_files.list -M Data/HMM12 Data/Triphones_fm.list

Getting below error:

WARNING [-7391] StepAlpha: Forward/Backward Disagree -5295.431996/-5504.871553 in HERest
WARNING [-7324] StepBack: File Data/f1_gate.htk - bad data or over pruning
in HERest
WARNING [-7391] StepAlpha: Forward/Backward Disagree -4466.774482/-4467.243444 in HERest
WARNING [-7324] StepBack: File Data/f1_sunroof.htk - bad data or over pruning
in HERest

Kindly help in solving the issue.

over 8 years ago

0 Yordan Kovachev over 8 years ago

TI__Guru**** 161600 points

Hi,

I've notified the sw team. Their feedback will be posted here.

Best Regards,
Yordan

0 Lalindra Jayatilleke over 8 years ago

TI__Mastermind 30365 points

Abhinandan,

Unfortunately, customization of the grammar and training models isn't supported.
We only provide the default models as reference to help you get started.

Lali

0 Abhinandan BR over 8 years ago in reply to Lalindra Jayatilleke

Intellectual 945 points

Hi Lali,

I trying to create a new set of models taking fresh voice recordings, segmenting them into phones, triphones, along with creation of filler model.

This is as per the guidelines given in TIesr_model_build/TIesr_Tools/Example/ReadMe.txt

ReadMe.txt does say the step by step procedure.
I don't understand why this is not possible? Please re-clarify this.

0 Abhinandan BR over 8 years ago in reply to Abhinandan BR

Intellectual 945 points

Hi Lali,

Kindly request you to confirm this as soon as possible.

0 Abhinandan BR over 8 years ago in reply to Lalindra Jayatilleke

Intellectual 945 points

Hi Lali,
I am creating new model set for a total of 10 words to be exact. For _fil model I am using a dictionary pronuncitaion of most possible words.
Is it feasible to create for a less words of number 10? Is this the cause for error?

Processors

Processors forum

TMDX5535EZDSP: TI Embedded Speech Recogniser training set generation