You trained how precisely ?
I am using below command to train it …
python3 DeepSpeech.py --export_dir /home/XXXXXX/proto_1/model1/ --train_files ./train.csv --dev_files ./train.csv --test_files ./train.csv --train_batch_size 1 --dev_batch_size 1 --test_batch_size 1 --epochs 75 --dropout_rate 0.15 --n_hidden 2048 --lm_alpha 0.75 --lm_beta 1.85 --learning_rate 0.0001 --lm_trie_path /home/XXXX/proto_1/model1/trie --lm_binary_path /home/XXXXX/proto_1/model1/lm.binary --alphabet_config_path /home/XXXXXX/DeepSpeech/data/alphabet.txt
I tried with changing below parameters-
test batch size- 1/2/3
dev batch size- 1/2/3
train batch size- 1/2/3
learning rate - 0.001/0.0001/0.0001/0.00001
Note : i am using 10 audio clips that’s why i tried with lower batch sizes
Well, you just have not enough data. You would need much more than that.
i am training , validating and testing on same audio clips , so far that as well ?
Well, not really, but with this dataset, you cannot really do anything serious, so it does not matter a lot 
1 Like
mate I’m currently downloading 16 gigs of German audio files to create a model, 10 clips aren’t enough.
1 Like