Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.

Pepper and Watson Speech related API

625 views

Published on

This is failure story of making translation function on Pepper.

Published in: Technology

Pepper and Watson Speech related API

  1. 1. Pepper and Watson Speech related API (a.k.a Failure story of making Translation App on Pepper) Forex Robotics Co., Ltd Copyright ©2015 – 2016 Forex Robotics Co. Ltd. All rights Reserved.
  2. 2. At first, This is failure sotry of development, _| ̄|○ Please take easy to listening …  Copyright ©2015 – 2016 Forex Robotics Co. Ltd. All rights Reserved.
  3. 3. Copyright ©2015 – 2016 Forex Robotics Co. Ltd. All rights Reserved. ML Research and Development Marketing prediction system Robot Development Robot App and systems Fin-tech Development MT4 EA App and systems Kaz Takahashi Forex Robotics CO., Ltd. CEO 1+1 1 Pepper, 1 Person (as of May 2016) Previous Developer and researcher Trend Micro Inc, about consumer product and security 2015 Established since Garage Entrepreneurs IBM Join Global Entrepreneur Program Get award of Pepper related hackathon 2times Certified Official Robo App Partner (Basic) Bluemix use Node-RED Watson API DashDB Official member Robot Revolutio Initiative
  4. 4. Copyright 2015 Forex Robotics Co. Ltd. Allright Reserved. Introduced Communication Robot Industory Map by Robot Start sorce:Communication Robot Industy Map / 2016 Q1 / Japan robot start inc.
  5. 5. Main Issue: One day, One customer ask me that … “Can Pepper translate between Japanese and Chinese without additional device?” Copyright ©2015 – 2016 Forex Robotics Co. Ltd. All rights Reserved.
  6. 6. Customer Insight •My customer’s store already sets up Pepper. •The store come many Chinese customer at one time. (By sight seeing bus, guess around 50 people) •But only 2 persons about Chinese speaker staff. •If Pepper translate between JP  Chinese, JP speaker staff may support Chinese customers. •No additional cost, because my customer already have bought Pepper. Copyright ©2015 – 2016 Forex Robotics Co. Ltd. All rights Reserved.
  7. 7. However, Copyright ©2015 – 2016 Forex Robotics Co. Ltd. All rights Reserved. Speech to text Text to speechLanguage Translation Pepper’s speech recognition APIs are not realized “free word” recognition. (need to define wording) Are Pepper + Watson API able to do that?
  8. 8. But the reality was not so sweet … Copyright ©2015 – 2016 Forex Robotics Co. Ltd. All rights Reserved. 音声認識 Speech to text 音声合成 Text to speech テキスト翻訳 Language Translation EN Portuguese Spanish French Arabic English (UK) Portuguese (Brazil) English (US) Japanese Chinese (北京語) Arabic Spanish Spanish English (US,UK) Portuguese (Brazil) French German Italian Japanese Functions don’t connect. ( ゚Д゚)
  9. 9. TO make matters worse… Pepper’s microphone is easy to pick up noise • Conjecture • Pepper has microphone on head top. (there are 4 microphones) • Also Pepper has CPU fan on head top, so easy to pick up the noise. • The sound data has disadvantage for Watson Speech Recognition API. Copyright ©2015 – 2016 Forex Robotics Co. Ltd. All rights Reserved. It should be silent.
  10. 10. So, from Pepper is Difficult to recognition speech by Watson API • 千 売り場は どこで • うん 無理は どこ • チェン氏は どこです • 支援 おりはどこ で • D_エー売り場とか • 遅延 売りは どこで • チェーン 売り場 とこ です • D_エー売り は どこ です Copyright ©2015 – 2016 Forex Robotics Co. Ltd. All rights Reserved. (例) 「チェーン売り場は どこですか?」 e.g. “Where is car chain section?”
  11. 11. To begin with, What is the speech recognition? (in Japanese) Copyright ©2015 – 2016 Forex Robotics Co. Ltd. All rights Reserved. O H A 1. Identified vowels and consonants based on spike. 千 チェン 遅延 チェーン 2. Analogy words based on identified sounds. • Conjecture • When pickup wording, I guess that selected high frequency appearance word. • If so, terminology is low rate to pick up, because low frequency appearance wording. • In the first place, there is possibility of not listed wording .(e.g. special terminology or coined word)
  12. 12. Summary • Pepper’s Speech recognition APIs are difficult about free word recognition. • Watson Speech recognition API is support free word recognition and grate support function for Pepper. • However, Japanese  Chinese translation function is not realize only Watson API. (as of May 2016) • Request of Watson Speech Recognition API • Need frequency control and add word function for terminology. • It's super, if improve recoginition by noises! Copyright ©2015 – 2016 Forex Robotics Co. Ltd. All rights Reserved.
  13. 13. Finally, I made proto type!  • I’ll do demo at Robot Forum 2016 in Forex Robotics booth. (July 1st, Aug 2nd) Copyright ©2015 – 2016 Forex Robotics Co. Ltd. All rights Reserved. http://youtu.be/tTufpC5xReo
  14. 14. At the end, Copyright ©2015 – 2016 Forex Robotics Co. Ltd. All rights Reserved. Wanted business opportunity about cooperation of the robot and IT systems! Also wanted partner company! (no matter JP or not) Boost up robot industry together! ktakahashi@forexrobotics.jp https://www.facebook.com/forexrobotics.jp/ Feel free contact me!
  15. 15. Thank you for your attention! Copyright ©2015 – 2016 Forex Robotics Co. Ltd. All rights Reserved.

×