Skip to content

Latest commit

 

History

History
38 lines (30 loc) · 2.25 KB

README.md

File metadata and controls

38 lines (30 loc) · 2.25 KB

Speech Recognition

This project is an stt module based on PocketSphinx

For hello_world program, pocketsphinx takes as input an audio file with following requirements

Record an audio file: signed 16-bit, little-endian, sample rate 16kHz, single channel
arecord -r 16000 -f S16_LE -c 1 -D plughw:0,0 -t raw test_hw.raw
Play an audio file: signed 16-bit, little-endian, sample rate 16kHz
aplay -f S16_LE -r 16000 test_hw.raw

In order to compile json_parser and json_test
gcc json_parser.c -I/usr/include/uuid -I/usr/include/json-c -ljson-c -o json_parser

To do the demo for lamp control
./voice_commandor -inmic yes -lm model_with_keyword/model_kwd.lm -dict model_with_keyword/model_kwd.dic -keyphrase "OBAMA" -kws_threshold 1e-5 -logfn log.log

1st lamp
./dummyLamp -a 224.0.29.200 -p 1234 -u 1b51f38e-ae97-41d0-a08d-297f68ed0be9

2nd lamp
./dummyLamp -a 224.0.29.200 -p 1234 -u 9b6b19b3-a213-4bee-865a-dba40a4472e1

Location tags are configured in device_config.json

Source Code Tree

  • ad_acous_module and ad_acous_module are two trained acoustic modules. They are of objective to show how to train an acoustic module. The difference is in ad_acous_module, I added 6 times the same phrase "turn on the light" in order to improve the recognition rate for the word "light". But as a result, it didn't seem to improve a lot.
  • json_test contains some practical examples for creating json object or parsing json object.
  • model_with_keyword is the final language model we trained and used since "portes ouvertes".
  • ss contains the screen shots of system performance while our program is running on a raspberry pi v3.
  • .asoundrc is an example for ALSA configuration.
  • .project is the project file under Eclipse.
  • corpus_file.txt is the corpus file that we used to construct the final language model.
  • device_config.json is the device list we mentioned in the technical report.
  • dummyLamp.c is the dummyLamp application developped by xAAL team.
  • dummy_commander.c is the interaction module.
  • smtc_module.c is the semantic analysis module.
  • stt_module.c is the STT module.
  • smtc_module.h contains the data structures and interfaces shared between semantic analysis module and interaction module.