Question on the BiRNN function of DeepSpeech

DeepSpeech
#1

I am using the deepspeechv0.01 for some test, and I want to ask is the batch size in this function must be 1?

Because when I tested to feed the function with a tensor (shape = (2, 47, 494)), there is a error :
InvalidArgumentError (see above for traceback): len(seq_lens) != input.dims(1), (1 vs. 2

[[Node: bidirectional_rnn/bw/ReverseSequence = ReverseSequence[T=DT_FLOAT, Tlen=DT_INT32, batch_dim=1, seq_dim=0, _device="/job:localhost/replica:0/task:0/device:GPU:0"](Reshape_5, qq_frames/read/_77)]]

[[Node: logits/_87 = _Recvclient_terminated=false, recv_device="/job:localhost/replica:0/task:0/device:CPU:0", send_device="/job:localhost/replica:0/task:0/device:GPU:0", send_device_incarnation=1, tensor_name=“edge_1559_logits”, tensor_type=DT_FLOAT, _device="/job:localhost/replica:0/task:0/device:CPU:0"]]

and when I feed the the function with a tensor (shape = (1, 47, 494)), it works.

#2

Do you mean Deep Speech v0.1? Is there some reason you are using such an old version?

#3

I’ve solved this issue, because my another parameter length didn’t feed correctly.

No special reason for using it.