PnG Speech Recognition
The Pick Up & Go (PnG) recognizer is enabled when the Asr_Engine_EngineName parameter is set to PnG.
- This is speaker-independent speech recognition.
- Pick Up & Go recognition requires no user training, so users can start right away.
If a user encounters problems using Pick Up & Go, they should retrain the words or consider additional ways to Improve Speech Recognition.
PnG FEATURE LICENSING
An Enterprise Voice or VoiceConsole license is required to use Pick Up & Go in languages other than American English. If it is not properly licensed, VoiceCatalyst uses the BlueStreak recognizer. Contact your customer service representative.
- If the A700x has Internet network access and a Cloud Licensing Service configuration file, the Cloud Licensing Server is used to license the device for Pick Up & Go.
- If the A700x has Internet network access but does not have a Cloud Licensing Service configuration file, VoiceConsole must be used to license the device for Pick Up & Go.
- If the A700x device does not have Internet network access, VoiceConsole must be used to license the device for Pick Up & Go.
The license indicates what features are being licensed, such as All Languages, Honeywell Voice for Manhattan Active, Pick Up & Go, etc.
Availability
The Pick Up & Go recognizer is available only on the A700x and requires VoiceCatalyst 4.2 or greater for American English and 4.3 or greater for other supported languages.
Languages Supported for Pick Up & Go
| Language | PnG <languageTag> |
|---|---|
| American English | en_US |
| Arabic (World Wide) | ar_WW |
| Arabic (Gulf) | ar_AE |
| Australian English | en_AU |
| Belgium Dutch | nl_BE |
| Brazilian Portuguese | pt_BR |
| British English | en_GB |
| Bulgarian | bg_BG |
| Canadian French | fr_CA |
| Chinese - Cantonese (Hong Kong) | zh_HK |
| Chinese - Mandarin (China) | zh_CN |
| Chinese - Mandarin (Taiwan) | zh_TW |
| Chinese - Sichuanese (China) | zh_CN |
| Czech | cs_CZ |
| Danish | da_DK |
| Dutch | nl_BE |
| Finnish | fi_FI |
| French | fr_FR |
| German | de_DE |
| Greek | el_GR |
| Hebrew | he_IL |
| Hungarian | hu_HU |
| Indian English | en_IN |
| Indonesian | id_ID |
| Italian | it_IT |
| Japanese | ja_JP |
| Korean | ko_KR |
| Latin American Spanish | es_MX |
| Malay | ms_MY |
| Netherlands Dutch | nl_NL |
| Norwegian | no_NO |
| Persian (Iran) | fa_IR |
| Polish | pl_PL |
| Portuguese | pt_PT |
| Russian | ru_RU |
| Slovak | sk_SK |
| Spanish | es_ES |
| Swedish | sv_SE |
| Thai | th_TH |
| Turkish | tr_TR |
Language Selection
PnG uses the same language as the text-to-speech voice that is loaded onto the device. It also uses the task or VAD’s phonetic substitutions for the loaded language.
Speaker-Independent Pronunciations
PnG expects the user to pronounce text the way a native speaker of the loaded language would pronounce it. For example, it would expect a different pronunciation for “male” when English is loaded than when Spanish is loaded.
When specified, the phonetic substitution display string is used to generate the pronunciation that the recognizer is looking for instead of the actual vocab word. This is for language translations and apps like VIO that map generic command tokens to real words. (The display string is used for both display and speech-in/recognition while the “substitution” or pronunciation string is used for speech-out/text-to-speech.)
The recognizer generates pronunciations automatically. If a pronunciation does not work well enough for one or more users, it can be replaced or supplemented with additional pronunciations using the PnG_Recog_PronunciationRespellings_<languageTag>_<word> parameter.
Trainability/Speaker-dependent Pronunciations
VoiceCatalyst 4.5 or greater enables a user to train speaker-dependent templates to help resolve poor recognition. PnG does not require training, however, configuration parameter settings can force enrollment training. VoiceCatalyst 4.5 or greater also enables the user to initiate retraining of a word via the button menu as was always available with BlueStreak).
Enrollment Training
In enrollment training, VoiceCatalyst prompts the user to say each word in the app before using the app. Either the PnG_Train_RequireAllSDTemplates parameter or PnG_Train_RequiredSDWords parameter can be set to enable PnG Trainability, forcing users to train. This is useful when users have a different dialect than that of the recognizer.
Update Training
VoiceCatalyst 4.5 or greater features the “retrain word” option on the button menu. During training, PnG prompts for four examples of each word during. (In contrast, BlueStreak prompts for ten examples and/or embedded word sequences.) There may be delays when using the app immediately after retraining.
Only the words at the current task node are available to update train. In order to update train a word, the operator must get to a point in the task where the vocabulary word is available to be recognized, and then initiate update training.
If the retrain word > all numbers option is on the button menu, BlueStreak speech recognition is being used. This option is not available for PnG.
To determine if PnG or BlueStreak speech recognition is being used, check for the All Numbers option at Button Menu > Retain Word > All Numbers. This option is not available for PnG, so if an All Numbers option is found, you are using BlueStreak.
There is no embedded training with PnG. It does not use the EMB file.
Both update and enrollment training take less time with PnG than with BlueStreak because there are fewer PnG prompts.
PnG Templates
For PnG, the template names that display in VoiceConsole contain the language tag in addition to the word. An example is American English (en_US). Templates trained in one language are not interoperable and are not loaded when other languages are being used.
PnG for US English generates templates with the extension PnG1. Other languages use template extension PnG2. PnG also creates encrypted PnG0 templates during training, which are identical to their corresponding PnG1 or PnG2 templates, except that they contain the audio examples recorded while the user trained. These templates:
- Ensure forward compatibility of trained templates with future versions of the recognizer.
- Are available for Honeywell support needs.
To remove the pronunciations that a trained or adapted template provides, the PnG0, PnG1, and PnG2 templates for a particular word for a particular language must be deleted from VoiceConsole. The operator profile must also be removed by loading to the devices the same or another operator before they are used.
If the operator profile is not reloaded to the device before being used, adaptation may occur, causing the adapted previous templates to be re-uploaded to VoiceConsole.
When a template is trained or adapted, the device adds the new pronunciations to the built-in pronunciations. Both pronunciations are used from that point forward. This differs from BlueStreak, which always replaces a template/pronunciation with a newly trained one. If the same PnG template is trained again, the new template replaces the previously trained template and the new pronunciations are again added to the built-in pronunciations.