Configuration file

In this page, a complete description of the configuration file required by the VASR (VASR - C++) engine is provided.
This table explains what are the different field for, their range of possible values (if application) and their default value (if any).

Note that on Android the log/<name>/sink and log/<name>/sink_options are ignored

Field path	Optional	Value type	Range	Default value	Description
`paths`	✔	Object	N / A		Object that contains the different path required by the engine to load specific resources
`paths/data_root`	✔	String	N / A	`.`	Base location where the engine will load resources from. If relative, it will be based on the configuration file base path
`paths/acmod`	✔	String		`acmod`	Directory where acoustic models (`.vam`) will be loaded from. If relative, it will be based on the `data_root` path
`paths/graph`	✔	String		`graph`	Directory where compiled grammars (`.vgg`) will be loaded from. If relative, it will be based on the `data_root` path
`paths/log`	✔	String		`log`	Directory where log will be placed. If relative, it will be based on the `data_root` path
`log`	✔	Object	N / A		Object that contains all the logging configuration of VASR
`log/<name>`	✔	Object	N / A		Configuration for the logger `<name>`. Possible logger names are [`vasr` (the default logger), `perf` (the performance logger), `recognizer:<name>` (the recognizer named `name`), `model:<name>` (the model named `name`)]. It it also possible tu use wildcard as specific places to configure multiple logger at once. Here are the allowed values: [``, `recognizer:`, `model:*`]
`log/<name>/level`	✔	String	[`trace`, `debug`, `info`, `warning`, `warn`, `error`, `err`, `critical`, `off`]	`info`	Set the minimal log level for this logger. Note that setting this will print the level specified and everything above the level e.g. setting the level to `warning` will still print `error` and `critical` messages. Of course `warn` and `warning` / `err` and `error` are equivalent
`log/<name>/sink`	✔	String	[`stdout`, `stdout-color`, `stderr`, `stderr-color`, `file`]	`stdout-color`	Set the destination of log messages for this logger
`log/<name>/sink_options`	✔	Object	N / A		Contains all options specific to the type of sink
`log/<name>/sink_options/path`	✔	String			(For `file` sink type only) Set the output file path. If relative, it will be based on the `paths/log` path
`log/<name>/sink_options/max_size`	✔	Int	[1'024 - inf]		(For `file` sink type only) Specify the maximum file size (in kbytes) before rotating the file
`log/<name>/sink_options/max_file`	✔	Int	[1 - inf]		(For `file` sink type only) Specify the maximum number of log file that will be kept
`log/<name>/sink_options/truncate`	✔	Boolean		false	(For `file` sink type only) If set to true file will be truncate / rotate (depending if the maxsize / maxfile options have been specified or not) when the engine is created
`asr`	✘	Object	N / A		Object that contains all the configuration of VASR
`asr/recognizers`	✘	Object	N / A		List here all recognizer that will be created during run-time
`asr/recognizers/<name>`	✘	Object	N / A		Specify the definition of a recognizer. The name is completely free and will be used in the code to create the Recognizer object
`asr/recognizers/<name>/acmods`	✘	Array	N / A		A list of all the path to the acoustic model that will be loaded and used by this recognizer. If relative, it will be based on the `paths/acmod` path
`asr/recognizers/<name>/settings`	✔	Object	N / A		Additional settings that will be passed to every acoustic model of this recognizer
`asr/recognizers/<name>/settings/min_speech_duration`	✔	Int	[100 - 10'000]	500	Amount of time (in milliseconds) that speech needs to be detected before triggering a recognition. Note that after a successful trigger the X ms of audio of this value will also be passed on to the recognizer. Needs to be a multiple a 100
`asr/recognizers/<name>/settings/min_silence_duration`	✔	Int	[100 - 10'000]	700	Amount of time (in milliseconds) that silence (non-speech) needs to be detected before triggering the end of the recognition. Needs to be a multiple a 100
`asr/recognizers/<name>/settings/padding_size`	✔	Int	[0 - 1'000]	450	Amount of perfect silence (in milliseconds) that will be artificially added to the audio at its beginning and its end. This improve performance at almost no cost. Need to rarely be modified.
`asr/recognizers/<name>/settings/speech_probability_threshold`	✔	Float	]0.0 - 1.0[	0.5	Set the sensibility of the VAD. Lower value mean more prone to false prositive and higher value to false negative
`asr/recognizers/<name>/settings/intermediate_result_frequency`	✔	Int	0 or [450 - 10'000]	750	Amount of audio (in milliseconds) that needs to be send to the engine before it returns an intermediate result. A value of 0 completely disable intermediate result. Note that this value will only be an aproximation
`asr/models`	✘	Object	N / A		List here all compiled grammars that will be loaded during run-time
`asr/models/<name>`	✘	Object	N / A		Specify the name of the model. The name is completely free and will be used in the code to load the appropriate grammar
`asr/models/<name>/file`	✘	String			Path to the compiled grammar file (`.vgg`) for this model. If relative, it will be based on the `paths/graph` path
`asr/models/<name>/settings`	✔	Object	N / A		Additional settings that will be passed to the model
`asr/models/<name>/settings/max_hypothesis`	✔	Int	[1 - 10]	3	Set the number of hypothesis that will be returned by the engine for this model

Example

JSON

{
  "paths": {
    "data_root": "../data",
    "acmod": ".",
    "graph": "."
  },
  "log": {
    "vasr": {
      "level": "off"
    },
    "perf": {
      "level": "info",
      "sink": "stdout"
    },
    "recognizer:rec": {
      "sink": "stderr-color"
    },
    "model:cmd": {
      "sink": "file",
      "sink_options": {
        "path": "cmd.log",
        "max_size": 10240,
        "max_file": 5
      }
    }
  },
  "asr": {
    "recognizers": {
      "rec": {
        "acmods": [
          "eng-US.vam"
        ],
        "settings": {
          "min_speech_duration": 500,
          "min_silence_duration": 700,
          "padding_size": 450,
          "speech_probability_threshold": 0.5,
          "intermediate_result_frequency": 0
        }
      }
    },
    "models": {
      "cmd": {
        "file": "eng-US.vgg",
        "settings": {
          "max_hypothesis": 5
        }
      }
    }
  }
}