lionsoul2014/freeswitch-asr

TTS and ASR module with auto Voice Active Detecting supported for Freeswitch. I build it for Nature sound interactive, With the embedded LUA engine we could easly build a Freeswtich application like this.

Description

TTS and ASR module with auto Voice Active Detecting supported for Freeswitch. I build it for Nature sound interactive, With the embedded LUA engine we could easly build a Freeswtich application like this.

How to use it ?

1. Copy and merge the ${PROJECT_ROOT}/src and the ${PROJECT_ROOT}/conf with the Freeswitch source tree.
2. Add the following two module in the ${FREESWITCH_SOURCE_ROOT}/modules.conf

asr_tts/mod_yytts
asr_tts/mod_yyasr

3. Re-compile and install the freeswitch to install mod_yytts and mod_yyasr modules.
4. Active the mod_yytts and the mod_yyasr by add the following to lines into the ${FREESWITCH_INSTALLATION_ROOT}/conf/autoload_configs/modules.conf.xml

<load module="mod_yytts"/>
<load module="mod_yyasr"/>

5. Copy the lua scripts under ${PROJECT_ROOT}/scripts to ${FREESWITCH_INSTALLATION_ROOT}/scripts/
6. Bind a number to build application by adding the following xml settings to the ${FREESWITCH_INSTALLATION_ROOT}/conf/dialplan/default.xml

<!-- yytts tts module testing -->
<extension name="yytts_demo">
  <condition field="destination_number" expression="^5001$">
    <action application="answer"/>
    <action application="sleep" data="2000"/>
    <action application="lua" data="yytts_demo.lua"/>
  </condition>
</extension>

<!-- yyasr asr module testing -->
<extension name="yyasr_demo">
  <condition field="destination_number" expression="^5002$">
    <action application="answer"/>
    <action application="sleep" data="2000"/>
    <action application="lua" data="yyasr_demo.lua"/>
  </condition>
</extension>

How to test it ?

Start the Freeswitch and install a Linphone client.

1. Dial 5001 for TTS testing, check ./scripts/yytts_demo.lua for detail usage, module interface overview:

session:answer();   -- answer the call

-- set to use the yytts TTS with zhilingf speaker
session:set_tts_params("yytts", "zhilingf");

-- fire the speak
session:speak("Hello，我是人工智能电话系统, Bye!");

-- hangup the call
session:hangup();

2. Dial 5002 for ASR testing, check ./scripts/yyasr_demo for detail usage, ASR interface overview:

session:answer();   -- answer the call

-- register the sound input callback
session:setInputCallback(function onInput(s, type, obj) end);

-- start to detect the speech to keep firing the feed interface insite yyasr module
session:execute("detect_speech", "yyasr directory directory");

-- resume the detect_speech
session:execute("detect_speech", "resume");

-- stop the detecting
session:execute("detect_speech", "stop");

-- check ./script/yyasr_demo for details usage.

Secondary development

TTS module

1. TTS module structure:

*module_interface = switch_loadable_module_create_module_interface(pool, modname);
tts_interface  = switch_loadable_module_create_interface(*module_interface, SWITCH_SPEECH_INTERFACE);
tts_interface->interface_name = "yytts";
tts_interface->speech_open = yytts_speech_open;
tts_interface->speech_close = yytts_speech_close;
tts_interface->speech_feed_tts = yytts_speech_feed_tts;
tts_interface->speech_read_tts = yytts_speech_read_tts;
tts_interface->speech_flush_tts = yytts_speech_flush_tts;
tts_interface->speech_text_param_tts = yytts_text_param_tts;
tts_interface->speech_numeric_param_tts = yytts_numeric_param_tts;
tts_interface->speech_float_param_tts = yytts_float_param_tts;

2. Invoke process for TTS module in Freeswitch:

open
while ( thread running ) do
  feed
  read
  flush
end
close

3. Embedded the other TTS engine by override the yytts_speech_feed_tts

static switch_status_t yytts_speech_feed_tts(switch_speech_handle_t *sh, char *text, switch_speech_flag_t *flags)
{
	// feed the wave stream to the yytts->audio_buffer buffer
	return SWITCH_STATUS_SUCCESS;
}

ASR module

1. ASR module structure:

asr_interface = switch_loadable_module_create_interface(*module_interface, SWITCH_ASR_INTERFACE);
asr_interface->interface_name = "yyasr";
asr_interface->asr_open = yyasr_asr_open;
asr_interface->asr_load_grammar = yyasr_asr_load_grammar;
asr_interface->asr_unload_grammar = yyasr_asr_unload_grammar;
asr_interface->asr_close = yyasr_asr_close;
asr_interface->asr_feed = yyasr_asr_feed;
asr_interface->asr_resume = yyasr_asr_resume;
asr_interface->asr_pause = yyasr_asr_pause;
asr_interface->asr_check_results = yyasr_asr_check_results;
asr_interface->asr_get_results = yyasr_asr_get_results;
asr_interface->asr_start_input_timers = yyasr_asr_start_input_timers;
asr_interface->asr_text_param = yyasr_asr_text_param;
asr_interface->asr_numeric_param = yyasr_asr_numeric_param;
asr_interface->asr_float_param = yyasr_asr_float_param;

2. Invoke process for ASR module in Freeswitch:

open
while ( thread running ) do
  feed
  check_results
  get_results
end
optional pause|resume
close

3. Embedded the other ASR engine by override the yyasr_asr_feed

/*! function to feed audio to the ASR */
static switch_status_t yyasr_asr_feed(
    switch_asr_handle_t *ah, void *data, unsigned int len, switch_asr_flag_t *flags)
{
	// feed the text to the yyasr->text_buffer text buffer
	return SWITCH_STATUS_SUCCESS;
}


/*
 * Voice Active detecting implementation.
 * Optimize the VAD by modify stop_audio_detect function.
*/
static switch_bool_t stop_audio_detect(
    switch_asr_handle_t *ah, int16_t *data, unsigned int samples)
{
    return SWITCH_FALSE;
}