CTC decoder | C++ implementation | Python implementation
2020-05-22: Now both input shapes [batch_size x timesteps x num_classes] and [timesteps x batch_size x num_classes] are supported.
The following code-skeleton gives a first impression of how to use the decoding algorithm with Python. More details can be found in the python_demo dir.
import numpy as np
from ctc_beam_seach_decoder_pywraper import ctc_beam_search_decoder
# Inputs' shape must be : timesteps x batch_size x num_classes
inputs = np.array([
[[0.6, 0.0, 0.0, 0.4, 0.0, 0.0 ]],
[[0.0, 0.5, 0.0, 0.5, 0.0, 0.0 ]],
[[0.0, 0.4, 0.0, 0.6, 0.0, 0.0 ]],
[[0.0, 0.4, 0.0, 0.1, 0.0, 0.5 ]],
[[0.0, 0.5, 0.0, 0.5, 0.0, 0.0 ]],
]).astype(np.float32)
# sequence_length shape must be [batch_size]
sequence_length = np.int32([inputs.shape[0]])
# decoded : [top_paths, batch_size, max_timestep]
# log_probabilities : [batch_size, top_paths]
decoded, log_probabilities = ctc_beam_search_decoder(inputs, sequence_length,
beam_width=50, top_paths=10)The following code-skeleton gives a first impression of how to use the decoding algorithm with C++. More details can be found in the cpp_demo dir.
#include "src/ctc_beam_search_decoder.h"
// Prepare Inputs
const int max_time = 5;
const int batch_size = 1;
const int num_classes = 6;
const int beam_width = 50;
const int top_paths = 10;
float inputs[max_time][batch_size][num_classes] = {
{{0.6, 0.0, 0.0, 0.4, 0.0, 0.0 }},
{{0.0, 0.5, 0.0, 0.5, 0.0, 0.0 }},
{{0.0, 0.4, 0.0, 0.6, 0.0, 0.0 }},
{{0.0, 0.4, 0.0, 0.1, 0.0, 0.5 }},
{{0.0, 0.5, 0.0, 0.5, 0.0, 0.0 }}
};
int sequence_length[batch_size] = {max_time};
// Prepare Outputs
int decoded[top_paths][batch_size][max_time];
float log_probabilities[batch_size][top_paths] = {{0.0f}};
int status = -1;
status = ctc_beam_search_decoder( &inputs[0][0][0],
sequence_length,
beam_width,
top_paths,
max_time,
batch_size,
num_classes,
&decoded[0][0][0],
&log_probabilities[0][0]);