1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
|
require 'libspeech'
local TNetReader = nerv.class("nerv.TNetReader", "nerv.DataReader")
function TNetReader:__init(global_conf, reader_conf)
self.feat_id = reader_conf.id
self.frm_ext = reader_conf.frm_ext
self.feat_repo = nerv.TNetFeatureRepo(reader_conf.scp_file,
reader_conf.conf_file,
reader_conf.frm_ext)
self.lab_repo = {}
for id, mlf_spec in pairs(reader_conf.mlfs) do
self.lab_repo[id] = nerv.TNetLabelRepo(mlf_spec.file,
mlf_spec.format,
mlf_spec.format_arg,
mlf_spec.dir,
mlf_spec.ext)
end
self.global_transf = reader_conf.global_transf
end
function TNetReader:get_data()
local res = {}
local frm_ext = self.frm_ext
local step = frm_ext * 2 + 1
local feat_utter = self.feat_repo:cur_utter()
local expanded = nerv.CuMatrixFloat(feat_utter:nrow(), feat_utter:ncol() * step)
expanded:expand_frm(nerv.CuMatrixFloat.new_from_host(feat_utter), frm_ext)
local rearranged = expanded:create()
rearranged:rearrange_frm(expanded, step)
local input = {rearranged}
local output = {rearranged:create()}
self.global_transf:init()
self.global_transf:propagate(input, output)
expanded = nerv.CuMatrixFloat(output[1]:nrow() - frm_ext * 2, output[1]:ncol())
expanded:copy_fromd(output[1], frm_ext, feat_utter:nrow() - frm_ext)
res[self.feat_id] = expanded
for id, repo in pairs(self.lab_repo) do
local lab_utter = repo:get_utter(self.feat_repo, expanded:nrow())
res[id] = lab_utter
end
self.feat_repo:next()
return res
end
|