Xingyu Na Senior Speech R&D Engineer, Apple Mailing Address: No. 2 Kexueyuan South Street, Beijing, China Office: Raycom Tower A Email: asr.naxingyu -at- gmail.com A printable version CV is here.

I received my Ph.D. degree from Beijing Institute of Technology in 2014 under supervision of Prof. Jingming Kuang and Prof. Xiang Xie. I was a visiting Ph.D. student in Dr. Philip N. Garner's group at Idiap Research Institute in 2012 and 2013.

News

07/09/2020: I'm joining Apple!
• 30/03/2020: Our book "Speech Recognition with Kaldi" is available on JoyBuy, Amazon and DangDang
• 18/09/2017: Our paper "AIShell-1: An Open-Source Mandarin Speech Corpus and A Speech Recognition Baseline" is accepted by Oriental COCOSDA 2017 as an oral presentation!
• 21/08/2017: I'm joining Microsoft China!
• 02/12/2016: I'm joining Alibaba Robotics Corp. as a Senior Staff Engineer!
• 05/09/2016: Attending Interspeech 2016 at San Francisco!
• 10/12/2015: I'm joining Letv as a Senior Researcher on speech recognition.
• 19/03/2015: Our paper "Incremental Syllable-Context Phonetic Vocoding" was accepted by TASLP.
• 14/07/2014: I gave a talk about "Real-Time Speech Synthesis" at IEEE ICME 2014.
• 07/03/2014: I'm joining Chinese Academy of Sciences, Institute of Acoustics as an Assistant Researcher.

Projects

Playing with Kaldi (the most popular open-source speech recognition toolkit)
My contributions are

• created components for convolutional neural network in nnet2
• created and tuned left-biphone setups for Chain model
• modified transition model and HMM topology kernel
• maintainer of aishell, fisher_swbd, hkust, gale_mandarin and thchs30 benchmarks
Playing with HTS (the most popular open-source speech synthesis toolkit)
I used HTS a lot for my PhD thesis. I shared the tools that were useful to me on Github and some of them are included as HTS extentions

• HTS_PDFparser: a lite parser for hts_engine model
• StreamGenerator: a single stream speech parameter generator for customizable hts_engine
• MGETraining: HTS training scripts supporting minimum-generation-error training
Voxforge for Chinese Calling for voluntary participants!

Publications

AISHELL-2: Transforming Mandarin ASR Research Into Industrial Scale
Jiayu Du, Xingyu Na, Xuechen Liu, Hui Bu
arXiv, 1808.10583, 2018 [paper] [Kaldi recipe]

AISHELL-1: An Open-Source Mandarin Speech Corpus and A Speech Recognition Baseline
Hui Bu, Jiayu Du, Xingyu Na, Bengu Wu, Hao Zheng
O-COCOSDA, Seoul, R. O. Korea, 2017 [paper] [Kaldi recipe]

Purely Sequence-trained Neural Networks for ASR based on Lattice-free MMI
Daniel Povey, Vijayaditya Peddinti, Daniel Galvez, Pegah Ghahrmani, Vimal Manohar, Xingyu Na, Yiming Wang, Sanjeev Khudanpur
Interspeech, San Francisco, US, 2016 [paper] [Kaldi recipe]

An Emperical Exploration of CTC Acoustic Models
Yajie Miao, Mohammad Gowayyed, Xingyu Na, Tom Ko, Florian Metze, Alexander Waibel
IEEE Conference on Acoustic, Speech and Signal Processing, Shanghai, China, 2016 [paper] [Eesen recipe]

Two-stage ASGD Framework for Parallel Training of DNN Acoustic Models using Ethernet
Zhichao Wang, Xingyu Na, Yonghong Yan
IEEE Automatic Speech Recognition and Understanding Workshop, Arizona, US, 2015 [paper]

Incremental Syllable-Context Phonetic Vocoding
Milos Cernak, Phil Garner, Alexandros Lazaridis, Petr Motlicek, Xingyu Na
IEEE/ACM Transactions on Acoustic, Speech and Language Processing, 23(6), 2015 [paper] [project]

Syllabic Pitch Tuning for Neutral-to-Emotional Voice Conversion
Lakshmi Saheer, Xingyu Na, Milos Cernak
Idiap Research Report, Martigny, Switzerland, 2015 [paper]

Low-Latency Parameter Generation for Real-time Embedded Speech Synthesis System
Xingyu Na, Xiang Xie, Jingming Kuang
IEEE International Conference on Multimedia And Expo, Chengdu, China, 2014 [paper]

Improving Voice Quality of HMM-based Speech Synthesis Using Voice Conversion Method
Yishan Jiao, Xiang Xie, Xingyu Na, Ming Tu
IEEE Conference on Acoustic, Speech and Signal Processing, Florence, Italy, 2014 [paper]

Syllable-based Pitch Encoding for Low Bit Rate Speech Coding with Recognition/Synthesis Architecture
Milos Cernak, Xingyu Na, Phil Garner
Interspeech, Lyon, France, 2013 [paper]

Convolutional Pitch Target Approximation Model for Speech Synthesis
Xingyu Na, Phil Garner
Idiap Research Report, Martigny, Switzerland, 2013 [paper]

An Improved Tone Labeling and Prediction Method with Non-uniform Segmentation of F0 Contour
Xingyu Na, Xiang Xie, Jingming Kuang, Yaling He
IEEE International Symposium on Chinese Spoken Language Processing, Hongkong, China, 2012 [paper]

Tone Generation by Maximizing Joint Likelihood of Syllabic HMMs for Mandarin Speech Synthesis
Xingyu Na, Chaomin Wang, Xiang Xie, Jingming Kuang, Yaling He
Speech Prosody, Shanghai, China, 2012 [paper]

Professional Activities

• Reviewer:
• Speech Communication
• EURASIP Journal on Audio, Speech, and Music Processing
• KSII Transactions on Internet and Information Systems
• IEEE Signal Processing Letters
• Journal of the Audio Engineering Society
• Pattern Recognition Letters

Github Activities