14.6 Speech recognition network sample

This section introduces the sample including speech separation, recognition, and success rate evaluation. Although the samples are for off-line use, you can use it for online processing just replacing AudioStreamFromWave to AudioStreamFromMic . All sample files are in Recognition directory. See Table 14.19 for details. The rest of this section describes how to run the samples step-by-step.

Table 14.19: The list of files

Category

File name

Description

Data

../MultiSpeech_en.wav

Wave file used in this sample

JuliusMFT

julius.jconf

Configuration file of JuliusMFT

 

AM/hmmdefs_en.bin

Acoustic model of English pronunciation

 

AM/allTriphones_en

List of triphones in the acoustic model

 

LM/order.*

Grammar-based language model

HARK

Recognition_en.n

HARK network file for localization, separation, and feature extraction

 

Recognition.sh

Shell script to run the network file

 

../config/microcone_tf.zip

Transfer function (for HARK 2.1.0 or later)

 

sep_files

Directory for separated sounds

Evaluation

score.py

Evaluation script

 

transcription_A.txt

Reference data of the utterances for each direction

 

transcription_B.txt

Reference data of the utterances for each direction