** Customized Large-Scale Datasets ** Olewave delivers customized, labeled, and validated large-scale real-world NLP/CV/speech/multimodal datasets of various scenarios such as dictation and conversation in multi accents/dialects/languages, and of diverse topics such as education, finance, legal, entertainment, healthcare, retail, and customer service. Our datasets include: • topic-specific text datasets for training your own LLM model • visual/video/image datasets with tags/prompts for training your own CV model • speech/audio datasets of different languages and dialects for training your own ASR/Whisper/SeamlessM4T/TTS model. • and multimodal datasets.We constantly collect timely data from languages including Latin America Spanish, Arabic, Southeast Asian, Chinese, Japanese, Korean.1/3-1/10 in data pricing vs. traditional data vendors with the same or better data efficacy.** Proprietary/Private Data Cleaning and Labeling **Olewave offers bespoke solutions for proprietary data labeling, normalization, and transformation.Tired of inaccurate transcriptions and frustrating APIs? We offer a superior solution with:• AI-powered Accuracy: Transcribe any audio, regardless of language, dialect, accent, or topic, with exceptional accuracy. We surpass the competition in understanding even the most challenging recordings.• Detailed Insights: Gain valuable insights with word/character-level confidence scores, precise timestamps, and advanced speech analytics.• Privacy Guaranteed: Keep your data secure. Integrate our powerful data labeling tool directly into your platform, eliminating risks associated with external APIs.• Competitive Pricing: Enjoy high-quality service at accessible prices, outperforming both tech giants and human-intensive transcription solutions.Ready to experience the difference? Don't settle for mediocrity. Contact info@olewave.com and give us a try!

To PhD students who are interested in doing research on but not limited to Seq2Seq-based large vocabulary continuous speech recognition (RNN-T) and spoken language understanding (End-to-End-based Conversational AI) with me, feel free to send your CV/resume to wei.w.chu@gmail.com.I have supervised 10+ interns, most of them ended their internship with publications, launched project, and systematic ways of doing academic research.

1) Full-stack development and improvement of acoustic, pronunciation, language modeling, and decoder optimization for streaming non-native large vocabulary continuous speech recognitiona. Acoustic Modeling (Research & Development): i. Multi-Task Learning on a limited amount of training data collected from each task. ii. Using Adversarial training to train for dictation task using read speech iii. Research on speaker/accent-embeddings and neural network topologiesb. Pronunciation Modeling/Dictionary (Research & Development): i. Automatically detect mispronunciation then generate a pronunciation dictionary for English learners.c. Language Modeling and Decoding (Development) i. Push the accuracy (10-15%) and latency (0.3-0.4s) of a streaming decoder to their limit. ii. Adding confidence score and domain-special re-scoring for PARTIAL hypothesis.2) End-2-End Speech Recognition: a. Compared the wordpiece vs character on hybrid CTC-Attention based network. Paper published.3) Video Synthesis and Smoothing a. Fast Audio/Text-driven Talking Head Synthesis on Limited Amount of Training b. Automatically smooth head movement when stitching video clipstag: kaldi, openfst, language model, acoustic model, pronunciation, deep learning, machine learning, speech synthesis, voice conversion, hotword detection, spoken keyword spotting, keyword recognition, DSP, speech activity detection, audio classification, automatic closed captioning, music composition, speech recognition, natural language processing, spoken dialogue system, computer vision, computer graphics, lip sync, talking head animation, noise robust speech recognition, deep neural network, recurrent neural network, convolutional neural network, AI, LSTM, End-to-End speech recognition, DNN, RNN, CNN, CTC, CUDA, GPU, cluster, hadoop, mapreduce, big data..

As the first Speech Research Scientist and Engineer at Snap Research, promptly prototype, build and customize speech recognition services, keyword spotting in audio services, and audio/speech processing libraries for product teams.1. Robust Closed Captioning for Audio Content-- Worked with News/Ads/User Content teams to customize the above speech recognizer to generate closed captions for sound tracks with music and noise.2. Spoken Keywords Spotting-- Co-developed and deeply customized a BLSTM-based keyword spotting engine on the cloud for curse word detection and brand name detection in noisy oral speech.3. Hotwords Detection on Mobile-- Worked with colleagues to deliver a DNN-based light-weight hotwords detector on iOS. Evaluated noise robust features. Co-invented a way of automatically generating data with transcriptions for training.3. Audio Signal Processing on iOS/Android\\-- Worked out purely DSP-based speech detection and enhancement libraries.-- Evaluated and improved audio fingerprint algorithm (Shazam-like). -- Designed and built a PSOLA-based spoken voice to singing voice conversion process (Patent filed). -- Worked with our intern on a RNN-based music composition machine that can automatically generate pleasant MIDI music with multiple instruments. Paper accepted. 4. Acoustic Event Detection-- Worked out a pipeline to extracting data for deep learning purpose. Co-tried different network structures. Customize it as a service for engineering teams.5. Deep Learning-based Voice Conversion-- Co-worked out a many-to-one BLSTM-based voice conversion algorithm. Paper submitted.6. Audio-driven Lip Sync-- Co-worked out a DNN-based audio driven lip sync algorithm which is well working and runs fast. Paper published.7. Multi-modal Emoji Prediction-- Co-worked out a deep learning-based Emoji predictor given audio and text. (Patent filed).8. Consulting for Internal Teams-- Audio Transformation (Voice Filter), Beamforming, Audio quality testing.

-- Built and delivered production-level Deep Neural Network (DNN)-based models for speech recognition.-- Built a low-cost but yet high-performance CPU+GPU computing cluster for deep learning experiments.-- Designed and implemented new practical algorithms for improving the performance of speech recognition.-- Handled practical aspects of a production-level speech recognizer, which include: * Speech and text normalization and lexicon modeling for training multi-language recognizers, * Improving a recognizer's robustness to speaker and environmental variations by performing adaptation, * Customization and optimization for speech recognition on different tasks and platforms, * Towards reliablely rejecting background noises and unrelated speech input to a recognizer, * Designing workflow and preparing data for validation and QA purpose.

Learning the skills of handling issues from vendors, customers, and internal develop and debug teams. Designed and built a test system for internal teams, customers and vendors. Learned how a corporation operates.

Understood the key aspects in a successful Interactive Voice Response (IVR) system. Also learned how a middle-sized company operates.

Earned 500k funding for the company. Also learned how a startup operates.

Audio-visual-depth automatic speech recognition on Kinect.

Worked out very accurate noise robust pitch estimation and tracking algorithm.Pioneer study in bird song classification, recognition, and detection.

Micrphone array-based kid speech recognition..

Statistically-based Letter-To-Sound conversion.

Discriminative training for speech recognition.

Acoustic event detection.

Speaker segmentation and clustering.