Moataz El Ayadi Email & Phone Number

Milton, On, Ca

Toronto, Ontario, Canada

Toronto, Canada Area

Technical leader of Operational Risk Data Science team - Led teams of Data Scientists and Data Scientists co-ops towards delivering ML-based solutions that satisfy the operational risk business requirements.Supervised and contributed to the following projects:- Semantic Analysis of Legal Contracts using BERT models- Sensitive Records Tagging using Named Entity Recognition technologies- Risk Categorization using Hierarchical Classification and Multitask Learning

Toronto, Canada Area

Lead of 'AI for Speech Insights' group:* Designed, implemented, and productionized an end-to-end voice-to-text engine with the following functionalities: voice activity detection, dial tone detection, speech/non-speech discrimination, speaker segmentation (using variational Bayes methods with i-vector), and speech recognition (using conventional senone-based DNN/HMM architecture with n-gram language model).Achievements:* Developed two services: 1. BITSY (Behavioural Intelligence and Transcription SYstem): which applies artificial intelligence and machine learning to surface actionable business insights from speech. 2. TOD (Transcription on Demand)* Delivered useful business insights for two clients: 1. Contact Call Centre 2. Scotia Life Insurance

Montreal, Canada Area

As a speech tech. consultant at TandemLaunch I am responsible for1. Improving the product ability to reject unknown or irrelevant commands.2. Developing a powerful voice activity detection algorithm.3. Guiding the team towards most recent research and technology directions in speech recognition.4. Aiding the company in filing their patents.

Toronto, Canada Area

Analysis, implementation, and evaluation of algorithms to improve the performance of VoiceTrust's speaker recognition engine in disturbed environments.Design and implementation of audio processing and machine learning algorithms that serve speech and speaker recognition tasks.Accomplishments:1. Improving feature extraction for speaker recognition Conducted a detailed experimental study on the effect of speech features on the performance of the speaker recognition system. Considered several combinations of feature types (MFCC, LFCC, LPC, PLP), normalization methods (mean and variance), and recognition techniques (GMM, HMM, DTW). Achieved 20-30% relative improvement in the recognition performance.2. Fusion of speech features for speaker recognition Aggregated strong speech features in order to further improve the performance of the speaker recognition system. Achieved 20-30% relative improvement in the recognition performance.3. Speech enhancement for speaker recognition Implemented a powerful noise spectrum estimation technique. Developed an accurate voice activity detector (VAD).

Cairo, Egypt

1. Natural Language Call Routinga) Natural Language Call Routing aims at saving the cost of human customer service and mobile network bandwidth by automatically understanding customer requests and routing them to the correct department.b) In this project, I implemented efficient text search algorithms for NLCR that achieved 90.5% and 91.67% correct recognition accuracies for Arabic and French text, respectively.2. Dynamic Pricing of Mobile Callsa) Dynamic pricing is the process of understanding, anticipating, and influencing consumer behaviour in order to maximize revenue from a fixed perishable resource (network bandwidth)b) The objective of this project is to guiding mobile subscribers towards more efficient use of the mobile network through incentives in the call tariff.c) Successfully developed statistical models to characterize the subscribers’ call behavior, e.g. call durations and rate of incoming and outgoing calls. d) Employed the obtained statistical models in developing powerful pricing strategies which results in a more efficient usage of the mobile networks and an increase in the revenue.

Cairo, Egypt

Teaching:Undergraduate courses:1. Probability and Statistics2. Partial Differentiation and Series Analysis3. Linear Algebra4. Complex Analysis5. Numerical AnalysisGraduate courses:6. Advanced Topics in Mathematics (random vectors, multivariate normal random variables, and stochastic processes)7. Pattern RecognitionResearchPerformance analysis of Fading Communication ChannelsSpeaker Identification under Noisy Environment

Conducted research on improving the performance on dimensionality reduction algorithms for pattern recognition problems.

1. Developed a new statistical model, vector autoregressive Gaussian mixture model (VARGMM) for adaptive text-independent speaker identification in noisy environment. 2. Applied the VARGMM to the problem of speech emotion recognition. The proposed classifier outperformed popular ones such as the hidden Markov model, the k-nearest neighbor, and the feed forward neural network.3. Applied VARGMM to the problem of equalizing multiple-input-multiple-output communication channel. The new equalization algorithm is compared to the whitening method and found to provide less symbol error probability and less relative error in the estimation of the channel state information (CSI) matrix.4. Developed a new fast and accurate method to calculate a tight upper bound for the error probability of the binary Gaussian classification problem. The proposed method is more accurate than the Chernoff bound and much faster than the Monte-Carlo simulation method.

Msc, Engineering Mathematics And Physics, Distinction With Honor

Phd, Electrical Engineering, A

Activities and Societies: University of Waterloo Egyptian student association (UWESA), vice president, Sep'06-Dec'07 Participated.