[FEATURE REQ] Cognitive Speech: Azure Maps search results emprovement and Speech, to engage Api

[gabser10]

AzureMaps Predictive Query prediction is missing addresses Asking for some addresses the related address is missing some address https://atlas.microsoft.com/search/address/json?subscription-key=XXXXXXXXX&api-version=1.0&query=via%20aurelia%20km%2013%20roma&countrySet=IT&&typeahead=true&language=%22it-IT%22 query	Result with maximum score (FreeFormAddress property in address object obrained by maps)
Via Aurelia chilometro/i /km 13 Roma	"Via Roma, 13, 17053 Laigueglia"
Via Antonio Marani 24 La Spezia	"Via Antonio Marani, 19123 La Spezia"(missing civic)
viale xxi aprile roma	"Via 25 Aprile, 22020 Pigra"
Viale 21 Aprile Roma	"Viale 21 Aprile, 21, 00162 Roma"
via xx settembre roma	"Viale 20 Settembre, 41049 Sassuolo"
Via 20 settembre roma	Via 20 Settembre, 20, 00187 Roma (adding civic)
via san pio decimo roma	Traversa Quinta Via San Pio Decimo, 89013 Gioia Tauro

Speech request to engage api results are frequently wrong Using Cognitive Speech recognition to recognize italian address (and proper name) .. predictions rarely improve with wrong query generated from speech recognizer, I'd need specifically for speaked address recognition api.

Thank you

[ghost]

Thanks for the feedback! We are routing this to the appropriate team for follow-up. cc @robch, @oscholz.

[oscholz]

We've passed this on to the owner of our Italian model, and that person will be looking into this, and comment here.

[fezuo]

For Cognitive Speech recognition request, tested with those AzureMaps query samples + clear speech, and recognition accuracy looks good. Could you let us know if there are some bug samples related to "results are frequently wrong Using Cognitive Speech recognition to recognize italian address" ?

[gabser10]

All the queries in table generate wrong results. query	Result with maximum score (FreeFormAddress property in address object obrained by maps)
Via Aurelia chilometro/i /km 13 Roma	"Via Roma, 13, 17053 Laigueglia"
Via Antonio Marani 24 La Spezia	"Via Antonio Marani, 19123 La Spezia"(missing civic)
viale xxi aprile roma	"Via 25 Aprile, 22020 Pigra"
Viale 21 Aprile Roma	"Viale 21 Aprile, 21, 00162 Roma"
via xx settembre roma	"Viale 20 Settembre, 41049 Sassuolo"
Via 20 settembre roma	Via 20 Settembre, 20, 00187 Roma (adding civic)
via san pio decimo roma	Traversa Quinta Via San Pio Decimo, 89013 Gioia Tauro

[fezuo]

As we tested Cognitive Speech Recognition request on above samples, with clear it-IT speech input got correct speech to text recognition result. To investigate your issue, could you share one or two speech samples so we can reproduce this issue?

[gabser10]

In the above examples are represented azure maps errors. In particular with the queries we obtain the right column result but is always wrong. I’ll attach Some audio with cognitive speech errors also.

[fezuo]

Thanks for clarification, please also let us know which one of AzureMaps Search error or Cognitive Speech Recognition errors is more critical to your usage.

[gabser10]

Our customers need to query for pratically all italian addresses in predctive mode (wiht typehad parameter setted to true). All the examples present in table are needed. The examples have in common Caesar's Numbers or Proper Name of people. The example "Via Antonio Marani 24 La Spezia" is particular also because we never obtain civic number in result. With Google Maps we obtain radically better results.

[fezuo]

I can help on Cognitive Service Speech to Text part. To improve entity name recognition, one suggestion is to use Custom Speech, it allows to upload batch of entity names such as street name, and improve recognition accuracy towards those names. Another ways is to improve the base model with specific common entity names, after the gap is identified. In addition, speech to text is enhancing it-IT entity names recognition and new model version will be updated by Dec.

[gabser10]

Thank you @fezuo, we know Custom Speech and could be usefull but very wasteful if we needs to train the model with all italian addresses. December could be very late for un update for customer needs. is there no way of joining Cognitive Speech and Azure Maps models?

Could be added The missing addresses in Azure Maps? And could this service enhances predictions in results?

[fezuo]

@gabser10 Thanks for the feedbacks. Our team can help to improve speech to text part. We need to verify and reproduce your issues in details. We only see special street name like "Via XX Settembre Roma" got issues in display format, that convert "XX" into 20 which seems wrong. We can correct this display format issue in short term. For other cases, please provide us expected speech recognition display results and audio sample if possible. The Italian model already support all common address, so we need to understand where is the gap. For prediction enhancement , do you mean speech to text recognition result or AzureMaps searching/ranking ?

[gabser10]

Thank you @fezuo I'll retrieve soon the audio needed to evidence Speech Problems, but The examples in table are relative to AzureMaps searching/ranking problems. query	Result with maximum score (FreeFormAddress property in address object obrained by maps)	Results problem
Via Aurelia chilometro/i /km 13 Roma	"Via Roma, 13, 17053 Laigueglia"	WRONG CITY AND CIVIC
Via Antonio Marani 24 La Spezia	"Via Antonio Marani, 19123 La Spezia"	missing street number
viale xxi aprile roma	"Via 25 Aprile, 22020 Pigra"	WRONG CITY
Viale 21 Aprile Roma	"Viale 21 Aprile, 21, 00162 Roma"	ADDING WRONG street NUMBER
via xx settembre roma	"Viale 20 Settembre, 41049 Sassuolo"	WRONG CITY
Via 20 settembre roma	Via 20 Settembre, 20, 00187 Roma	Adding wrong street number
via san pio decimo roma	Traversa Quinta Via San Pio Decimo, 89013 Gioia Tauro	Wrong city wrong street

[ghost]

Hi @gabser10. Thank you, for your interest in helping to improve the Azure SDK experience and for your contribution. We've noticed that there hasn't been recent engagement on this pull request. If this is still an active work stream, please let us know by pushing some changes or leaving a comment. Otherwise, we'll close this out in 7 days.

[ghost]

Hi @gabser10. There was a mistake and this issue was unintentionally flagged as a stale pull request. The label has been removed and the issue will remain active; no action is needed on your part. Apologies for the inconvenience.

[SaurabhSharma-MSFT]

Hi @gabser10 Can you please let us know if your issue has been resolved and this issue can be closed.

Thank you

[SaurabhSharma-MSFT]

Hi @gabser10 Following up again to know if you still need help on this issue.

[ghost]

Hi, we're sending this friendly reminder because we haven't heard back from you in 7 days. We need more information about this issue to help address it. Please be sure to give us your input. If we don't hear back from you within 14 days of this comment the issue will be automatically closed. Thank you!