how to predict results in Attention_tf_data_generator_STT?

[usamaehsan]

Aslam o Alaikum thanks for your great work can you please tell me how i can predict results in Attention_tf_data_generator_STT?

[masoudMZB]

Alaik-om-Aslam

If I undrestood well, you want to train with Attention model. and then predict one Sound.

For training you should first calculate MFCC or Spectrogram for sounds. then feed it to Network ( As in notebook )

for predicting you should first calculate MFCC or Spectrogram for sound. then use predictions = model.predict(your_sound_mfcc_or_spectrogram)

If there is still any question please let me know to answer you. thanks for reading this repository

[usamaehsan]

wa alaikum slam sorry for my dumb questions i am new to this feild and your repositry is only repositry on github which is helpfull for learning ASR even for a beginer like me. if i understood correctly model=model.load(saved checkpoint from training) right?

[usamaehsan]

@masoudMZB by the way if you will write code for predicting single sound in after training code in notebook it will be very helpfull for beginners and i will be very thankfull to you

[masoudMZB]

Asking questions is best way to upgrade your knowledge.

yes you are right. to clear it more it is like this prediction = your_last_checkpoint.predict(new_voice_to_transcribe)

If you want to see most of the codes available in github for ASR check our wiki page in this repo : https://github.com/shenasa-ai/speech2text/wiki/Audio-Prepossessing-and-Guide-for-this-Type-of-Projects

in this page I Wrote every thing I found for start in this field. good luck man

[masoudMZB]

@usamaehsan I will add predicting code soon. Thanks for your suggestion.

[usamaehsan]

thanks bro you are very helpfull

[masoudMZB]

your welcome. If you had any other idea share it with us.

I will close this issue.