Speech Recognition, Text to Speech, and Voice Interfaces
Download as PPTX, PDF1 like1,304 views
This document discusses speech recognition, text-to-speech, and voice interfaces. It describes speech recognition as allowing a user to speak commands into a device instead of pressing buttons. It then discusses uses of speech recognition for individuals with disabilities, medical transcription, dictation, and accessing menu commands. The document goes on to explain how speech recognition works by converting audio into recognized speech through steps like fast Fourier transforms and matching sounds to entries in a codebook. It also discusses discrete vs continuous speech recognition and the future of assistive technology in schools.
Speech Recognition, Text to Speech, and Voice Interfaces
- 1. Speech Recognition, Text-To-Speech, and Voice Interfaces By: Taryne Cahalin Stephanie Sirico Christiana Vasquez Adelphi University - Mobile Learning, Fall 2013
- 2. What is Speech Recognition? Instead of an automated voice recording that enables a person to press buttons, he or she is able to speak specific words into a device and command orders with the help of a speech recognition program.
- 3. The Uses Individuals With Disabilities – Assists those who have visual impairment, hand immobility, dyslexia, etc. Medical Transcription – Reduces delays to write out medical transcriptions Dictation - Converts words to text in emails or other word documents (also helpful for English Language Learners). Access Menu Commands – Opens files using voice commands.
- 5. How does it work? Speech recognition functions as a pipeline: The pipeline converts PCM (pulse code modulation) digital audio into recognized speech from a sound card.
- 7. Transforming PCM Digital Audio 16,000 PCM values per second, a “wavy line”, that repeat while the user speaks Information is converted for better recognition in the program Fast-Fourier transform identifies frequency components of a specific sound The program can approximate how our ears distinguish the sound
- 8. Transform PCM digital audio using Fast-Fourier Transform Fast-Fourier analyzes every 1/100th of a second and converts the audio data Each 1/100th produces an amplitude graph These graphs are in a database called a “codebook” Sounds matched to the most similar entry in the codebook. Sound is given a number which describes the sound, called the “feature number”
- 9. Two Categories Small Vocabulary/many-users: • Leaves room for speech disparity (i.e. accents) • Limited, preset number of commands that are able to be used Large Vocabulary/limited-users: • Best for business settings • Train system to work with a small number of users • Accuracy rate will increase as it learns its users
- 10. Discrete vs. Continuous Speech Discrete • Easier for program to understand • Noticeable pause after each word Continuous • Allows speaking at conversational speed • Used in most modern systems Programs now can recognize accents and pronunciations better. In earlier programs, accents, pronunciations, speed, and background noise were all variables that made sounds difficult for programs to understand.
- 11. Using Talk – Text to Voice This app allows you to type and then have the device repeat what was typed. In this case, instead of the device saying Taryne as “Ta-rin”, it pronounced it as “Ta-reen”. This is an example of speech recognition programs still need some work to be done because of emphasis on a syllable. The codebook did not have Taryne in it, so it was unable to pronounce her name.
- 12. The Future of Assistive Technology in Schools Students who need assistance in their writing skills because they have stronger oral skills. Students who were absent for a class, have poor memory, or need assistance hearing the lesson. Students who need assistance during Guided Reading. Students who are English Language Learners. Students with visual/hearing impairments and learning disabilities regarding reading/spelling/writing.