This file controls settings related to the speech server.
It is located by default in /etc/lumenvox/ on Linux and in C:\Program Files\LumenVox\Engine\config\ on Windows. See Configuration Files for more information about other configuration files.
The following parameters can be set. The format to use within the configuration file is PROPERTY_NAME = VALUE
[GLOBAL]
This section contains global configuration settings for the SRE (Speech Engine)
VERSION
Description: Contains information about the version of the software that created the configuration file. Do not modify this.
Possible Values: This should not be modified by users.
[SRE]
The SRE section controls the settings for the ASR speech engine.
SRE_PORTNUM
Description: The port number that the speech server will use to listen for incoming speech requests.
Possible Values: A valid port number.
Default Value: 5730
ENABLE_CACHE
Description: Whether grammar caching should be turned on or off. Grammar caching allows the server to automatically cache all loaded grammars to disk. This can save a lot of time when loading larger grammars, as they must be compiled before use, and this can take a while for big grammars.
Possible Values: 0 - 1
Default Value: 1
CACHE_EXPIRATION
Description: The amount of time, in minutes, the Engine will use a cached grammars on disk. If a cached grammar on disk is older than this value, the Engine will disregard the cached copy. If you are making extensive use of caching to preload very large grammars, you may want to set this very high.
Possible Values: An integer. If it is set to 0, the Engine will never used cached grammars.
Default Value: 1440 minutes (one day)
CACHE_MAX_NUMBER
Description: The maximum number of grammars to be kept in cache. The higher the number of grammars kept in memory, the more memory gets used in storing them.
Possible Values: An integer. If it is set to 0, the Engine will not cache any grammars.
Default Value: 1000
CACHE_MAX_MEMORY
Description: The maximum amount of memory, in bytes, that can be used to cache grammars
Possible Values: An integer. If it is set to 0, the Engine will not cache any grammars.
Default Value: 268435456 (256 MB)
NUM_DECODE_THREADS
Description: Maximum number of concurrent decode jobs that the ASR server will run. Increasing this number makes it more likely that parallel decode tasks must compete for resources. Decreasing this number makes it more likely that a decode task must wait in a backlog before being processed.
Possible Values: An integer. If it is set to 0, the Engine will not perform any decodes.
Default Value: 4
FRAME_TRACK_MODE
Description: Determines how available resources are used to track information required for confidence scoring
Possible Values: 0 - 2. If set to 0, confidence scores will be poor but memory use will be lowered and decode speed will be increased. If set to 1, confidence scores will be good, memory use will be medium, but decode speeds will be slow. If set to 2, confidence scores will be good, memory use will be high, and decode speeds will be medium. In most cases, you should set this to either 1 or 2 depending on how you value speed versus memory use.
- 0 - No frame tracking (fastest, least memory but worst confidence scoring)
- 1 - History_lists (slowest, less memory)
- 2 - History blocks, (slightly faster, more memory)
Default Value: 2
CRITICAL_MEMORY_THRESHOLD
Description: When the system running the speech server falls below this many available MB of memory, the speech server will enter a critical resource mode. This causes it to serialize all decode tasks, attempt to free up pooled memory and grammars, and disables frame tracking (resulting in worse confidence scores).
Possible Values: An integer representing the minimum amount of free system memory (in MB) you wish to have before the system enters critical resource mode.
Default Value: 100
LOW_MEMORY_THRESHOLD
Description: When the system running the speech server falls below this many available MB of memory, the speech server will enter low resource mode. This causes it to reduce the number of decode tasks it will handle in parallel.
Possible Values: An integer representing the minimum amount of free system memory (in MB) you wish to have before the system enters low resource mode.
Default Value: 250
LIMITED_MEMORY_THRESHOLD
Description: When the system running the speech server falls below this many available MB of memory, the speech server will enter limited resource mode. This causes the ASR to reduce the number of decode tasks it will handle in parallel and limit frame track mode to 0 or 1.
Possible Values: An integer representing the minimum amount of free system memory (in MB) you wish to have before the system enters limited resource mode.
Default Value: 750
LOAD_XXX_RES_MODEL
Description: Determines which of the three resolution levels (LOW, MED, HIGH) of acoustic model are loaded. Combinations (LOW and HIGH) can be enabled, but the lower resolution will be defaulted to. These are part of an experimental effort to improve accuracy. Ultimately, we found the resource expenses of running the MED and HIGH models outweighed the gains and we recommend that users leave only the LOAD_LOW_RES_MODEL enabled (the default).
Possible Values: 0 - 1 (for each resolution)
Default Value: LOW enabled
LOAD_LOW_RES_MODEL = 1
LOAD_MED_RES_MODEL = 0
LOAD_HIGH_RES_MODEL = 0
FST_DETERMINIZE_SUBSET_LIMIT
Description: The maximum size a subset list can grow during FST determinization before it gives up. This value is a safeguard against determinization operations that can recurse infinitely or take an unreasonable amount of time and resources to complete.
Possible Values: A positive integer.
Default Value: 50000
LOGGING_VERBOSITY
Description: Controls the verbosity of event logging. This can be used to increase or decrease the amount of information logged by the application. Note that increasing the logging verbosity causes increase in CPU usage, and should therefore be avoided wherever possible in a production environment where optimal performance is critical.
Possible Values: 1 - 3
- 1 = Minimal logging. Logs only errors and critical issues.
- 2 = Medium logging. Logs all non-debug information, includes types covered in Minimal logging as well.
- 3 = Maximum logging. Logs all types of events. This will include any and all informational and debugging activity.
Default Value: 1
ENABLE_APP_STAT_LOGGING
Description: Updates statistics file when enabled.
Possible Values: 0 - 1
- 0 = Do not save any application statistic information.
- 1 = Save application statistic information.
Default Value: 1
Added in 17.0.200
OUT_OF_SERVICE
Description: Puts the ASR into Out-Of-Service mode where NEW decode and grammar load requests will be rejected, allowing maintenance. Any existing requests being processed will be allowed to continue to completion.
Possible Values: 0 - 1
- 0 = Normal operation mode.
- 1 = Out-Of-Service mode. New requests rejected.
Default Value: 0
PUBLIC_RULE_ACTIVATION_MODE
Description: This setting is used to control the SRGS compliance mode governing top-level grammar rules that are not referenced. Certain interpretations of the SRGS compliance specification require that these rules be enabled, however most users will want the potentially non-compliant option where such rules are not activated. Generally, most users should use the default setting unless you have a compelling reason not to.
Possible Values: 0 - 1
- 0 - Backwards compatibility mode: Rules that are not referenced are unreachable
- 1 - SRGS compliance mode: Top-level public rules that are not referenced are activated along with root
Default Value: 0
UNKNOWN_LANGUAGE_VALUE
Description: String value to use if unknown language identifier is specified in a grammar Set to "NULL" for VXML-compliant behavior (error on unknown language-id)
Possible Values: This needs to be a language identifier string (Ex: "AmericanEnglish", "ColombianSpanish") that corresponds to an acoustic model. See Language Identifier for the list of supported languages. Set to "NULL" for VXML-compliant behavior (error on unknown language-id)
Default Value: AmericanEnglish