Hi,
what are the output nodes of deepspeech? I’m looking at the old version 0.1.0.
I’m using this script to generate output_graph.pb from a checkpoint folder, containing .data, .meta, and index files.
import tensorflow as tf
#Step 1
#import the model metagraph
saver = tf.train.import_meta_graph('check.meta', clear_devices=True)
#make that as the default graph
graph = tf.get_default_graph()
input_graph_def = graph.as_graph_def()
sess = tf.Session()
#now restore the variables
saver.restore(sess, "check")
tf.compat.v1.disable_eager_execution()
#Step 2
# Find the output name
graph = tf.get_default_graph()
for op in graph.get_operations():
print (op.name)
#Step 3
from tensorflow.python.platform import gfile
from tensorflow.python.framework import graph_util
output_node_names = 'b1,b2,b3,b5,b6,bidirectional_rnn/bw/basic_lstm_cell/bias,bidirectional_rnn/bw/basic_lstm_cell/kernel,bidirectional_rnn/fw/basic_lstm_cell/bias,bidirectional_rnn/fw/basic_lstm_cell/kernel,h1,h2,h3,h5,h6'
output_graph_def = graph_util.convert_variables_to_constants(
sess, # The session
input_graph_def, # input_graph_def is useful for retrieving the nodes
output_node_names.split(",") )
#Step 4
#output folder
output_fld ='./'
#output pb file name
output_model_file = 'model1.pb'
from tensorflow.python.framework import graph_io
#write the graph
graph_io.write_graph(output_graph_def, output_fld, output_model_file, as_text=False)
The original output_graph.pb file (the one downloaded from the website) is 491.0MB and the one I generated is 490.9MB.
When I tried to transcribe my audio, this is what I got:
Loading model from file output_graph.pb
Loaded model in 0.221s.
Running inference.
Error running session: Not found: FeedInputs: unable to find feed output input_lengths
None
Inference took 0.006s for 1.975s audio file.
I’m not sure why, please help anyone?