International Women's Day 2023 yurikimura LT

Transcript

1. sisterに相談したら 人生が開けた話 Yuri Kimura “sister” Turned My Life
2. Self Introduction: Yuriって誰？ Love Python and Natural Language Processing Graduated from Temple University Japan Didn’t fit in Japanese high school → TUJ was soo comfortable to me!! My Dream is: Work in the company that has wide variety of people Make a care bot software like Baymax from Big HERO 6!
3. Started my career as a ML engineer in 2021, the midst of the Covid Disaster. Everyone working remotely Every tech event held online No environment around me to take the time to share my concerns and problems with someone. Then I met “sister”, which saved my life! How I met “sister”
4. What is “sister” Consultation and support platform for women working/aspiring in the IT industry! Met a lot of fascinating Sisters: - who has been in a management position in an IT company for more than a decade - who uses her unique strengths as a freelancer - who organizing a women's tech study community - who has completed a full English coding bootcamp and is an engineer in a foreign company…etc
5. Jumped into a coding-bootcamp: Le Wagon Data Science Course (9 weeks program) 7 weeks to cover areas related to DS and ML - Basic Python, SQL, Web Scraping - Mathematical Optimization with scipy - Statical Inference - Classic ML with sklearn - Deep Learning (MLP, CNN, RNN, Transformer) - Docker, MLflow, GCP(BigQuery, GC Run) Development of API as graduation project in 2 weeks
6. API is created in the last 2 weeks final project Our API 0:02-0:08 “John” 0:09-0:11 “Naomi” 0:12-0:20 “Hiro” 0:21-0:46 “Naomi” … Drop audio/video file Or youtube link Get the predictions with timestamp Our team made “Speaker Annotation API”
7. Data Preprocessing time “Speaker 1” input audio segments spectrograms MFCCs • MFCC reflects the way humans perceive audio. • Converted audio data into image data !
8. Model Training Input Output “Speaker 1” : 0.4 “Speaker 2” : 0.4 “Speaker 3” : 0.1 “Speaker 4” : 0.1 Layer #1 Layer #2 Dense • Input: Image data • Layer #1 and Layer#2 : Extract features from the input • Dense Layer: Classification • Output: Probabilities of who of the speakers the input image may correspond to Convolutional Neural Network (CNN) Check our Demo Video! ⬇ https://m.youtube.com/watch?v=n6CLCdD_JN o
9. Sum up and my future The insights I gained from my consultation with Sister turned my life around People cannot live alone. sisters' support saved my life. Now forming a career where I work hard at what I love and am good at: - Participation in Natural Language Processing Laboratory study group - Participation in Technical Book PJ Appreciative to sister, which was the kickoff of multiple turning points, and to the people who were involved in my life!