I will first Porvide what i have done in a detailed way:
my alphhabet.txt:
a b c d e f g h i j k l m n o p q r s t u v w x y z '
my vocabulary.txt:
jarvis
ive created lm.build with kenlm with the following commands
/lmplz --text vocabulary.txt --arpa words.arpa --o 3
/build_binary -T -s words.arpa lm.binary
ive create the tire from native client generate_tire.
my train and dev and test CSV Files are the same and they are the following:
wav_filename,wav_filesize,transcript
/home/shadi/Desktop/DeepSpeech-0.6.1/data/jarvis/train/jarvis.wav,327796,jarvis
my run.sh:
python3 -u DeepSpeech.py
–train_files /home/shadi/Desktop/DeepSpeech-0.6.1/data/jarvis/train/train.csv
–dev_files /home/shadi/Desktop/DeepSpeech-0.6.1/data/jarvis/train/train.csv
–test_files /home/shadi/Desktop/DeepSpeech-0.6.1/data/jarvis/train/train.csv
–alphabet_config_path /home/shadi/Desktop/DeepSpeech-0.6.1/data/alphabet.txt
–lm_binary_path /home/shadi/Desktop/DeepSpeech-0.6.1/data/jarvis/lm.binary
–lm_trie_path /home/shadi/Desktop/DeepSpeech-0.6.1/data/jarvis/trie
–dropout_rate
–epochs 200
–es_steps 10
–learning_rate 0.000025
–export_tflite true
–export_dir /home/shadi/Desktop/DeepSpeech-0.6.1/data/jarvis/results/model_export
–checkpoint_dir /home/shadi/Desktop/DeepSpeech-0.6.1/data/jarvis/results/checkout
Here is my problem after i take the output and run it with mic_vad steaming no matter what i say or input to the mic the result is that it is always detecting the words jarvis…
i know i dont have anywhere near of an amount of data but still if i give a totaly not matching sound or word how does match it to that word anyway?