Ai based character recognition and speech synthesis
Download as PPTX, PDF4 likes1,215 views
The document discusses an AI seminar on character recognition and speech synthesis. It describes how optical character recognition can convert scanned images or text into machine code, and speech synthesis can artificially produce human speech. It provides details on preprocessing techniques for character recognition, such as de-noising and binarization of images. It also explains the processes of text analysis, phoneme generation and prosody generation used in speech synthesis engines.
Ai based character recognition and speech synthesis
- 1. Seminar on “ AI Based Character Recognition and Speech Synthesis” Developed By: Kalyani Hadke Rani Kubetkar Shreya Surjuse Ankita Jadhao Kruttika Sorte Guided By Prof. H. N. Datir
- 2. Artificial Intelligence based Character Recognition and Speech Synthesis
- 3. NEED!!! We are facing so many problem in our daily life like, if we capturing the image some time we can not get proper image and not recognize the words. Lots of people have the problem of illiteracy . So we wish that this image should be converted to text for various purposes. While studying, we don’t read the text as a regular practice. So we wish that this text can be converted into audio. Apart which we wish should be captured in image & converted into audio. As generally we prefer hearing songs,
- 4. Introduction to CR and SS • Optical Character Recognition (OCR) is an electronic or mechanical converter. • OCR converts scanned images or text into machine code. • Speech Synthesis is the artificial production of human speech. • Speech synthesizer – a computer system used for this purpose. • TTS engine performs: • Language into speech • Symbolic linguistic representation to speech
- 5. • Image OCR • Recognized text TEXT • Speech engine speech •Image OCR •Recognized text TEXT• Recognized text TEXT • Speech engine speech Overview
- 6. DFD For Character Recognition System Pre-Processing Segmentation Image Digitization Network ImplementationTraining of Learning Network recognition Network testing Pre-processing explanation
- 8. Pre-Processing Segmentation Image Digitization Network ImplementationTraining of Learning Network recognition Network testing Pre-processing explanation DFD For Character Recognition System
- 9. Image segmentation Decompose sequence of characters in individual symbols. Directly affects the rate of recognition of script. Locate and identify boundaries of image. 1. External segmentation 2. Internal segmentation SEGMENTATION
- 10. . . Image segmentation is the process of partitioning an image into multiple segments ,so as to change the representation of an image into something that is more meaningful and easier to analyze. 1 2 3 4 . External Segmentation: determine the character lines in the text. Image segmentation is the process of partitioning 1
- 11. I m a g e Internal Segmentation: decompose an image of sequence of characters to images of individual symbols
- 12. Pre-Processing Segmentation Image Digitization Network ImplementationTraining of Learning Network recognition Network testing Pre-processing explanation DFD For Character Recognition System
- 13. • Mapping of symbol image into a corresponding two dimensional binary matrix • Issue – deciding the size of matrix • Sampling strategy for mapping the symbol image Image Digitization - Matrix matching
- 14. Input alphabet ‘ a ‘ 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 1 1 1 Segmented grid Digitization
- 15. • To feed matrix data to the network it must be linearize to a single dimension 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 1 1 1 …………...0 1 1
- 16. N A M E NAME 001110100…. 111010011…. 11001100…. 000111101….. NAME NEURAL NETWORK 14 1 13 5Image of scanned document Sub- images of individual letter from document Binary representation of sub-images. E.g 0 is white and 1 is black. A supervised neural network that has been trained to recognize images of characters. Neural network output numeric values corresponding to the recognized characters. File contains the text of the scanned document.
- 17. DFD For Character Recognition System Pre-Processing Segmentation Image Digitization Network ImplementationTraining of Learning Network recognition Network testing Pre-processing explanation
- 18. Artificial neural network consists of a large number of highly interconnected processing elements (neurons) working in unison to solve specific problems analogous to the biological neurons in the brain. Neurons communicated with weighted links NEURON NEURON Weighted link X1 Xn Output Wk1 Wkp Summation Sigmoid function
- 19. • Feed-forward neural network • A multilayer perceptron • Teaching and adaption of ANN • Implementation the ANN
- 20. Neural Network Input Signal Output signal Input layer First hidden layer Second hidden layer Output layer
- 21. DFD For Character Recognition System Pre-Processing Segmentation Image Digitization Network Implementation Training of Learning Network Recognition Network testing Pre-processing explanation
- 22. Neural Network Input Signal Output signal Binary converted image Obtained text of scanned image Back-propagation for Error calculation ERROR
- 23. N A M E NAME 001110100…. 111010011…. 11001100…. 000111101….. NAME NEURAL NETWORK 14 1 13 5 Sub- images of individual letter from document Binary representation of sub-images. E.g 0 is white and 1 is black. A supervised neural network that has been trained to recognize images of characters. Neural network output numeric values corresponding to the recognized characters. File contains the text of the scanned document. Image of scanned document
- 24. Speech Synthesis Pre-Processing Segmentation Image Digitization Network ImplementationTraining of Learning Network Input Image File containing Text of scanned document NLP DSP SPEECH TEXT TTS Engine
- 25. • TTS-Text to Speech engine • a computer-based system that read any text aloud. • TTS engine consist of Front-end - NLP Back-end -DSP Speech Synthesis
- 26. Modules of Text-to-Speech Natural language processing Text Preprocessing Text Analysis Linguistic Analysis Digital signal processing Speech Synthesizer TEXT SPEECH Prosody Phonemes Figure 1. A simple but general functional diagram of a TTS system Input Output
- 27. Speech Synthesis Pre-Processing Segmentation Image Digitization Network ImplementationTraining of Learning Network Input Image File containing Text of scanned document NLP DSP SPEECH TEXT TTS Engine TEXT ANALYSIS Auto PHONEME Prosody Generation
- 28. • This step called high-level, front-end or text- to-phoneme. • It consists of the following parts: Text analysis Automatic Phonetization Prosody generation NLP Module
- 29. Speech Synthesis Pre-Processing Segmentation Image Digitization Network ImplementationTraining of Learning Network Input Image File containing Text of scanned document NLP DSP SPEECH TEXT TTS Engine TEXT ANALYSIS Auto PHONEME Prosody Generation
- 30. NLP Module Text Analysis A pre-processing A morphological analysis A contextual analysis A syntactic-prosodic Text analysis
- 31. Speech Synthesis Pre-Processing Segmentation Image Digitization Network ImplementationTraining of Learning Network Input Image File containing Text of scanned document NLP DSP SPEECH TEXT TTS Engine TEXT ANALYSIS Auto PHONEME Prosody Generation
- 33. Speech Synthesis Pre-Processing Segmentation Image Digitization Network ImplementationTraining of Learning Network Input Image File containing Text of scanned document NLP DSP SPEECH TEXT TTS Engine TEXT ANALYSIS Auto PHONEME Prosody Generation Formant synthesis Concatenative synthesis
- 35. Speech Synthesis Pre-Processing Segmentation Image Digitization Network ImplementationTraining of Learning Network Input Image File containing Text of scanned document NLP DSP SPEECH TEXT TTS Engine TEXT ANALYSIS Auto PHONEME Prosody Generation Formant synthesis Concatenative synthesis
- 36. DSP component • Low level phoneme to speech • There are two main technologies used for the generating synthetic speech waveforms: • Concatenative synthesis • Formant synthesis
- 37. Speech Synthesis Pre-Processing Segmentation Image Digitization Network ImplementationTraining of Learning Network Input Image File containing Text of scanned document NLP DSP SPEECH TEXT TTS Engine TEXT ANALYSIS Auto PHONEME Prosody Generation Formant synthesis Concatenative synthesis
- 38. Formant Synthesis • Formant synthesis – rule-based synthesis • does not use any human speech samples at runtime. • Wave-form created using an acoustic model of the human vocal tract. • Generates artificial, somewhat robotic speech
- 39. Speech Synthesis Pre-Processing Segmentation Image Digitization Network ImplementationTraining of Learning Network Input Image File containing Text of scanned document NLP DSP SPEECH TEXT TTS Engine TEXT ANALYSIS Auto PHONEME Prosody Generation Formant synthesis Concatenative synthesis
- 40. Concatenative synthesis • Based on the concatenation of segments of recorded speech. • Gives the most natural sounding synthesized speech.
- 41. Concatenative Synthesis Diphone Concatenation Synthesis Unit Concatenation Synthesis Somewhat robotic speech, sonic glitches natural speech SUBTYPES
- 42. • Unit Concatenation Synthesis – Algorithm • Break language down to small units (phonemes, syllables, etc.) • Create a large database of recorded speech • Each unit is labeled: pitch, duration, prosody, position in syllable, etc. Labeling is synthesizer-dependant • Target utterance is selected at runtime by determining the best chain of units (HMM, Decision Tree) • Use DSP to smooth transitions between units Approaches To Wave-form Generation Concatenative
- 43. Pre-Processing Segmentation Image Digitization Network ImplementationTraining of Learning Network Input Image File containing Text of scanned document NLP DSP SPEECH TEXT TTS Engine TEXT ANALYSIS Auto PHONEME Prosody Generation Formant synthesis Concatenative synthesis
- 44. Advantages • Machine Language Translation • Information Retrievals • Visual Issue (Difficulty seeing text) • Motor Issue (Difficulty handling a book or paper)
- 45. QUESTIONS ????