白拾的笔记本

DiphonePhones are not a suitable unit for waveform concatenation, so we used diphones, which capture co-articulation. Diphone starts at the middle of one phone and ends at the middle of the other. Coarticulation is the overlapping of adjacent articulations or the influence of the target phoneme on surrounding phonemes. Middles of phones are more stable in their spectral properties than the edges, because of coarticulation. So, concatenating diphones should lead to smoother joins Waveform conca ...

Tokenisation & normalisation标签化 & 正则化 When processing almost any text, we need to find the words. This involves splitting the input character sequence into tokens and normalising each token into words. Handwritten rulesEvery user of a language holds a lot of knowledge about that language in their mind. One way to capture and make use of that knowledge is in the form of rules. Finite state transducerFinite State Transducers provide general-purpose machinery for rewriting an input sequen ...

Prerequisite knowledgestates of mobile robot, motion model, position orientation and velocity, More challenge see: GPS Kalman FilterPredict, measure, combining Prediction and correction Linear Kalman Filter Recursive Least Squares + Process Model Extended Kalman FilterLinear approximation, first-order term, still linear. Linearized motion model, Linearized measurement model. Jacobian matrix Limitation of Kalman Filter Summary The Kalman Filter is very similar to RLS but includes a motion mode ...

Problem Formulationpixel-level prediction Challenge of Semantic Segmentation General vision challenges: occlusion, truncation, scale and illumination changes Challenges specific to segmentation: smooth boundaries are hard to obtain due to the intrinsic ambiguity and resolution limitation in the image space. Segmentation with DNN Semantic Segmentation for Scene Understanding RANSAC algorithm boundaries Additional materialTrue Positive, False Positive, False Negative individual object help ...

Module 1: Phonetics and Representations of SpeechSystems in Speech ProductionSpeech production involves three systems in the body: the respiratory system, the phonation system, and the articulation system (Figure 1.2). The Respiratory SystemThe respiratory system supplies the air needed to initiate speech sounds (see Figure 1.3). It consists of parts of the body that allow us to breathe, including the lungs, the diaphragm, the muscles of the rib cage, and the abdominal muscles. The Phonation S ...

HarmonicsIn the frequency domain, periodic signals have harmonic structure: they contain energy only at multiples of their fundamental frequency. Voice sounds different from unvoiced sounds, has repeating pattern, in periodicity. So, the peak of the sound in the frequency domain is clear to observe. Impulse trainAn impulse train is the simplest periodic signal that has energy at all multiples of its fundamental frequency (and energy are evenly distributed). Spectral envelopeVarying the shape o ...

Introduction This week we continue the theme of looking at general kinds of harm technology can cause if we’re not careful with it. The topic we’re looking at is one that, with a few exceptions, gets a lot less press I think than bias because it’s a lot harder to quantify (and as established in week 1, computer scientists love quantitative data) That topic is human beings, and specifically what we might unintentionally affect when we remove people from a system, or at least change their involvem ...

Introduction We’ve covered some important background concepts in talking about Responsibility, Power and Data. The idea behind these has been to motivate some of why we should be interested in ethics in computer science, and why we are often going to be at least partially responsible for anticipating and mitigating harms that technical artifacts can bring about. For this middle portion of the course, we are looking at some broad categories of harm (some of the Data readings begun this process). ...

The increasing availability of digital data reflecting economic and human development, and in particular the availability of data emitted as a by-product of people’s use of technological devices and services, has both political and practical implications for the way people are seen and treated by the state and by the private sector. Yet the data revolution is so far primarily a technical one: the power of data to sort, categorise and intervene has not yet been explicitly connected to a social ju ...

Definition Your ability to see your will made manifest in the world. Explanation If you want to do something, how easy is it for you to do it. To naively put this in informatics terms, you could think of it as the weight associated with a person’s node in a network. Because we live in a world of systems (political, legal, cultural, organisational, social), a lot of that is tied up in how those system are designed. And a lot of that is dependent on who they were designed for, or by. Because, int ...

Challenge of object detectionnot fully observed, scale distraction, illumination changes. Basic conceptsbounding box and class labels, intersection of union (IoU) See more at: Evaluationg Mateics 2D Object Detection Steps (inference)feature extractor, computationally expensive, lower widthe and height, greater depth Prior bounding boxes, or anchor bounding boxes, assume bounding boxes, then guess where and how large they are. centroid location (where), box dimensions (size) Every pixel in ...

Since most material has been covered in previous blogs, I will go through this lecture contents in brief. Brief overview More ContentsActivation function example Task example For classification we have: Softmax output layer Cross-entropy Loss function For regression we have: Linear output layer Mean square error lLoss function Supplement reading LeCun, Yann, Yoshua Bengio, and Geoffrey Hinton. “Deep learning.” nature 521.7553 (2015):436-444. A Comprehensive Guide to Convolut ...

Image Features: A General Process Step 1 - Feature Detection: identify distinctive points in our images. We call these points features. Step 2 - Feature Description: associate a descriptor for each feature from its neighborhood. Step 3 - Feature Matching: we use these descriptors to match features across two or more images. Feature DetectionFeature Define Features: Points of interest in an image defined by its image pixel coordinates [u, v]. Points of interest should have the following ch ...

Introduction of computer visionComputer vision is a field of artificial intelligence (AI) that enables computers derive meaningful information from digital images, videos and other visual inputs obtained by a camera. Bucket of photons Photons converted to electrons Shift electrons along row for readout The readout on the device will translate the analog signal to either grayscale images or RGB images Image Formation pinhole camera, optical centre, focal length Stereo Cameras Image filter ...

Time domainSound is a wave of pressure travelling through a medium, such as air. We can plot the variation in pressure (captured by microphone) against time to visualise the waveform. Sound sourceAir flow from the lungs is the power source for generating a basic source of sound either using the vocal folds or at a constriction made anywhere in the vocal tract. somehthing about pressure with our vocal folds, the air flow is slow, its only the power source of sound, the pressure change is the ke ...