Niasche Aquino Email & Phone Number

• Working on Big Tech company TTS application. Reviewing/Improving guidelines for linguistic consultants; Providing and approving phonemic transcription of lexicon entries; Script proofreading and recording quality control. Model’s synthesis quality assessment and to propose overall improvements guidelines.• I conducted linguistic, phonological, and phonetic analyses to enhance the model’s overall performance and Brazilian Portuguese localization. I am responsible for addressing the main prosodic, phonemic, and morphosyntactic issues and solving them with the company’s proprietary tools.

• I researched and developed proprietary phonetic tools, resources, and speech technologies, which included TTS encoders, acoustic models and vocoders (improving overall quality, intelligibility, and naturalness), dialect-specific G2P converters, phonetically-rich datasets and algorithms for its construction, and ASR/TTS benchmarking tools as Phoneme Error Rate and naturalness evaluation algorithms.• I developed all resources using Python on Linux servers. I worked on state-of-the-art generative models (as Grad-TTS and BigVGAN) from the training dataset creation to fine-tuning and benchmarking. I implemented solutions to improve quality and phonetic representation of Brazilian Portuguese technologies.• To achieve commercial results, I used different NLP techniques and resources. For instance, in a text dataset construction tool, a greedy algorithm that encompasses text representation, lemmatization, PoS tagging, and RegEX for foreign words and prohibited grapheme sequence detection, G2P conversion, and triphones higher variability sentence selection.

• My research developed in cooperation with the National Organization of Public Security (PROCAD-SPCF20201812362P) produced new forensic technologies – funded by the National Council for Scientific and TechnologicalDevelopment and the São Paulo Research Foundation.• I developed a feature extraction algorithm in Praat Scripting Language for Speaker Identification with several local and prosodic parameters. The program has been used in national and international forensic cases.• I performed statistical tests and worked with machine learning algorithms to evaluate features’ noise robustness, minimum sample size, and discriminatory power.• Respectively, Kruskal-Wallis and Wilcoxon post-hoc tests for different signal-to-noise conditions. Changepoint analysis to estimate variance stabilization. Principal Component Analysis for dimensionality reduction and K-means clustering for unlabeled data based on the principal components. Classification modeling with Flexible Discriminant Analysis based on Multivariate Adaptive Regression Splines.

Research project in Acoustic Phonetics (modeling and speech analysis for automatic feature extraction tools).

Research project in Biolinguistics (modeling an evolutionary path for language).

Doctor Of Philosophy - Phd, Acoustic Phonetics - Signal Processing

Master'S Degree - Acoustic Phonetics, Analysis And Modeling Of Natural Language Sounds

Automatic acoustic features extraction for forensic speaker identification/comparison tasks. The statistical analysis performed were.

Bachelor Of Applied Science - Basc, Linguistics