Inference only returns one word

Archives DeepSpeech
1

Whenever I give an audio file, the correct first word is inferred and it skips the rest of the auio.

Doing transfer learning on urdu language

2

So, again, share context. There’s nothing we can do with that.

3

I created a lm and hence a scorer

for training:
!python3 deepspeech/DeepSpeech.py \

--drop_source_layers 1 \

--alphabet_config_path using_data/my_new_alphabets.txt \

--save_checkpoint_dir output_checkpoints2 \

--load_checkpoint_dir output_checkpoints2 \

--train_files   using_data/train.csv \

--test_files  using_data/test.csv \

--export_dir output_checkpoints2/model\

--log_dir output_checkpoints2/model\

--scorer_path deepspeech/data/lm/kenlm.scorer \

--n_hidden 2048 \

--noearly_stop \

--learning_rate 0.00095 \

--dropout_rate 0.22 \

--epochs 5

for inference:

!deepspeech --model output_checkpoints2/modeeellllllll/output_graph.pb --audio 001FCN017.wav --scorer deepspeech/data/lm/kenlm.scorer >> text_file.txt

4

It always writes one work (the first one ) ( correctly) even for 7sec audios.

5

@i140224_ma Please close this thread and SHARE ALL THE INFORMATIONS IN ONE THREAD ONLY, I’m getting really upset having to constantly switch between N threads to get a clear picture of your status.