cortana / speech platform - sec.ch9.mssec.ch9.ms/slides/winhec/2_03_cortana_speech_platform.pdf ·...

Post on 30-Jan-2018

223 Views

Category:

Documents

0 Downloads

Preview:

Click to see full reader

TRANSCRIPT

Cortana / Speech Platform

2

Cortana

truly personal

across all your devices get things done.

Cortana Across your Devices

Cortana Usage is Growing

In 9 months since Windows 10 launched

• Over 270 million active Windows 10 devices

• Over 5 billion questions asked of Cortana

• One million voice questions are directed towards

Cortana on a daily basis

Of these, 15% use the keyword

to initiate

On any given day, 25% of Cortana PC users

use speech

&Hey Cortana

Cortana Drives Customer Satisfaction

intent to recommend

improve customer satisfaction

increase “re-purchase” intent

With Windows 10 Anniversary Update

Nearly 1,000 Apps

withVoice

Commands

Cortana functionality for Windows 10 is available in the following languages and countries

English-US

English-UK

Chinese-China

German-Germany

French-France

Italian-Italy

Spanish-Spain

Geographies

Japanese-Japan

English-India

English-Canada

English-Australia

Portuguese-Brazil

French-Canada

Spanish-Mexico

In plan for Windows 10

Anniversary Update

Cortana with Voice (CwV) Marketing Assets

Icons only

Icons + Titles

Titles and

descriptions

Technical

Requirements

to market

Cortana with

Voice (CwV)

*The “Cortana with Voice” title, icon, description, and associated assets

such as messaging can only be used if all following criteria are met:

1. The device is tested using Microsoft’s “Speech platform tools”

2. The device’s test results meet or exceed the Standard spec

Cortana with Voice

Your truly personal digital assistant who helps you get things done, even

hands-free - just say “Hey Cortana” to get started.

Cortana with Voice

Standard vs. Premium Spec

Windows 10 Anniversary Update & Beyond

Standard • Normal ambient noise level

• 0.5m

• Standard experience

• Cortana with voice (CwV)

• Normal ambient noise level

• 0.5m

• Standard experience

• Cortana with Voice (CwV)

• Additional tests for Voice Activation

(optional)

Premium • More challenging conditions

• <1m• Talk level: 89 dBA• Premium experience

• More challenging conditions

• 4m• Talk level: 99 dBA• Far-field experience

• Additional tests for Voice Activation

*Early draft, subject to change

Cortana with Voice Testing Scenarios

Run the OEM Verifier tool and ensure the driver settings aligns with the design

• Utility - OEMVerificationx86.exe

• Driver reports these key elements correctly

Number of microphones

Microphone geometry

Driver mode supported

Follow the test guidance to validate Cortana with Voice performance

• Validate in 3 scenarios: Quite, Ambient Noise (café and pub), Echo

Parse data and ensure the test result are at or above the provided benchmark

• Utility - OEMScoreUtilityFarAndNearx64.exe

• Calculate the speech accuracy score & troubleshoot the issues

Cortana with Voice Validation Process

Audio effects supported

Exposes Audio Pipeline being used

Default microphone gain

Examples of Test Result

OEM Pipeline Microsoft Pipeline Speech Accuracy Score

Future Planning for CwV Requirements

*Early draft, subject to change

Platform Experience

• Platform providers test voice activation false accepts (FA) and correct accepts (CA) for every supported locale

• These two tests are not needed to be run by OEMs

Release People Scenario Near-field Far-field

0

degrees 50

degrees 0

degrees 50

degrees

2016

Male

Quiet (Private Office, Living Room)

90% 90%

Medium (Coffee Shop, Kitchen)

Music listening in Quiet

Female

Quiet (Private Office, Living Room)

90% 90%

Medium (Coffee Shop, Kitchen)

Music listening in Quiet

Future

Male

Quiet (Private Office, Living Room)

92% 92% 90% 90%

Music listening in Quiet

Medium (Coffee Shop, Kitchen)

Music listening in Medium Ambient

Female

Quiet (Private Office, Living Room)

92% 92% 90% 90%

Music listening in Quiet

Medium (Coffee Shop, Kitchen)

Music listening in Medium Ambient

Children (5-12)

Quiet (Private Office, Living Room)

92% 92% 90% 90%

Music listening in Quiet

Medium (Coffee Shop, Kitchen)

Music listening in Medium Ambient

OEM Experience

OEMs test specific devices for good voice activation accuracy on EN-US and JA-JP. These tests will become quality gates beyond 2016.

OEMs test specific devices for good

speech recognition on EN-US only.

When the Microsoft far-field processing is used the recommendation is to use the microphone array geometry shown above

For 360 degree operation an 8-mic circular array is recommended

For 3rd party solutions use their spec recommendations for microphone array and geometry

Recommended Array for Far-Field

Technical Requirement References

Requirement Detail Reference / Spec

Speech Platform

test tools

Application and

content guidance for

testing

Microsoft Speech Platform Test Tools

http://connect.microsoft.com/site1304/Downloads/DownloadDetails.aspx?DownloadID=59735

Speech Platform Audio Calibration files

http://connect.microsoft.com/site1304/Downloads/DownloadDetails.aspx?DownloadID=57500

v1.1 “Standard”

benchmark

Quiet, Ambient, Echo

scores

Speech Platform Input Device Recommendations v1.2

(§2 – Device.SpeechRecognition recommendations)

http://connect.microsoft.com/site1304/Downloads/DownloadDetails.aspx?DownloadID=58099

https://msdn.microsoft.com/en-us/library/windows/hardware/dn915051(v=vs.85).aspx

v1.1 Test setup Measurement setup Speech Platform Input Device Test Setup v1.1

http://connect.microsoft.com/site1304/Downloads/DownloadDetails.aspx?DownloadID=58099

https://msdn.microsoft.com/en-us/library/windows/hardware/dn915051(v=vs.85).aspx

Test setup for

accessories

Measurement setup Speech Platform Input Device Recommendations – Accessories Support v1.0

http://connect.microsoft.com/site1304/Downloads/DownloadDetails.aspx?DownloadID=59861

v2.0pre3 Specs Benchmarks and

measurement setup

Speech Platform Input Device Recommendations v2.0pre3

Speech Platform Input Device Test Setup v2.0pre3

https://connect.microsoft.com/site1304/Downloads/DownloadDetails.aspx?DownloadID=60086

• Ensure your drivers have the correct settings (mic geometry, mic gain, number of mics, expose audio pipeline being used)

• Download and test the latest specs and tools

• Provide feedback on Speech Platform v2.0pre3 specs

Call to Action

Online Survey Formhttp://aka.ms/winhecfeedback

Join WinHEC LINE Community @winhec

Download WinHEC presentations here:

https://channel9.msdn.com/Blogs/WinHEC* Gifts are limited. They will be offered at

“a first come, first serve” basis.

Please provide feedback on this session:

http://aka.ms/winhecfeedback

top related