Language to Language TranslationA Way to Homogeneous India... Mentor:Prof. K.T.Talele Team effort of:Anasree Chatterjee & Diwa Arunashree Why the system ?? What is language? Need for.
Download ReportTranscript Language to Language TranslationA Way to Homogeneous India... Mentor:Prof. K.T.Talele Team effort of:Anasree Chatterjee & Diwa Arunashree Why the system ?? What is language? Need for.
Language to Language TranslationA Way to Homogeneous India... Mentor:Prof. K.T.Talele Team effort of:Anasree Chatterjee & Diwa Arunashree Why the system ?? What is language? Need for proper communication Hazards of miscommunication Hence Need for our system Key users of our system Our system overview.... Any out of 8 languages Hindi English Enjoy the words ! Speak in one language & listen in another language in just 3 steps ! Speech to Text Text to Text Bengali text to Bengali speech Text to Speech English or Hindi speech to English or Hindi text e.g. English English text to text of selected output language e.g. Bengali Speech to Text Architecture ! Voice Input Phonetic Lexicon Analog to Digital Feature Extraction Language Model Acoustic Model Speech Engine/Decoder Store Word in a File 1.Voice Input 3.Feature Extracting 2.Analog to Digital Noise Filtering Speech to Text Architecture ! Voice Input Phonetic Lexicon Analog to Digital Feature Extraction Language Model Acoustic Model Speech Engine Store Word in a File Acoustic Model Components of ASR contd.... Audio Recording Tool CMU Sphinx Train Text Transcription Software Statistical Representations of the Sounds that make up each Word ACOUSTIC MODEL Uses Hidden Markov Model (HMM) Speech to Text Architecture ! Voice Input Phonetic Lexicon Phonetic Lexicon Analog to Digital Feature Extraction Language Model Acoustic Model Speech Engine Store Word in a File Components of ASR contd.... Phonetizer Valid words from output of acoustic model Phoneme -basic unit of Contains words + phonetic PHONETIC LEXICON Phonetic representation of every word in vocabulary Hindi :- Itrans-3 English :- phonetics Hindi Speech Sound Wave Hindi Script /UTF8 Itrans-3 IT3 to UTF8 Phoneme In:d:iyaa इंडिया / ইংডিযা / ఇండియా / ഇംഡിയാ/ இடியா Paanii पानी / পানী / పానీ / പാനീ / பான ீ Phoneme Hindi.dic Hindi Word In:d:iyaa इंडिया / ইংডিযা /ఇండియా /ഇംഡിയാ/ இடியா Paanii पानी / পানী / పానీ / പാനീ / பான ீ English Speech English word Pocket Sphinx Sound Wave Phoneme SphinxTrain Phoneme Cmu07.dic Word Speech to Text Architecture ! Voice Input Analog to Digital Feature Extraction Language Model Phonetic Lexicon Acoustic Model Speech Engine Store Word in a File Language Model Components of ASR contd.... Captures underlying grammatical structure of language. Statistical Language Model assigns probability to sequence of m words by probability distribution. Most common language models – n-gram LM LANGUAGE MODEL Tool CMUCLMTK USE:Restrict Word Search Steps of Language Model:Create Word frequencies CORPUS.TXT Vocabulary file Corpus N-gram file Language Model in .ARPA format CMU Cam LM TOOL KIT CORPUS.ARPA .ARPA File Speech to Text Architecture ! Voice Input Phonetic Lexicon Analog to Digital Feature Extraction Language Model Acoustic Model Speech Engine Store Word in a File Speech Engine Components of ASR contd.... Aspects of Speech Decoding Modified Version DTW Algorithm used Compares input speech data with acoustic models Determine which part of signal is speech and filter out silence durations Uses SPEECH ENGINE / DECODER Tool CMU Sphinx-PocketSphinx Samples of PocketSphinx acting as a Decoder.... Text to Text Architecture Retrieve Stored Word from File E.g. India FIND RETRIVE Database Script of Word in Selected Language E.g. इंडिया / ইংডিযা / ఇండియా / ഇംഡിയാ/ இடியா Use & Creation of Database! Text to Speech Architecture ! Phonetic Synthesizer Text parser Input Text in UTF8 Encodings Text to Phonetic Script Conversion Speech Synthesizer CV Pair Algorithm Sound concatenation Grapheme To Phoneme Rules Sound Database Speech Grapheme to Phoneme Conversion ! Phonetic description syllable based. 8 kinds of sounds allowed V: a plain vowel CV: a consonant followed by a vowel VC: a vowel followed by a consonant CVC: a consonant followed by a vowel followed by a consonant HCV: a half consonant, followed by a CV HCVC: a half consonant, followed by a CVC 0C: a consonant alone G[0-9]*: a silence gap of the specified length (typical gaps (C -consonant, V -Vowel, H-Half Sound) Consonants & Vowels ! CONSONANTS :- VOWELS :- Text to Speech Architecture ! Phonetic Synthesizer Text parser Input Text in UTF8 Encodings Text to Phonetic Script Conversion Speech Synthesizer CV Pair Algorithm Sound concatenation Grapheme To Phoneme Rules Sound Database Speech Text to Phonetic Script ! Unicode text common script. Speech Synthesizer common script Words in Hindi Examples Specific Phoneme (G2P) CV Pair (Pronunciation Sound) khana kh2 n2 CV CV maun m13n CVC kahaan k1 h2an CV CVC pratibha pHr1 t3 bh2 HCV CV CV sankalp s1n k1l 0p CVC CVC 0C chandramaa ch1n dHr1 m2 CVC HCV CV praan pHr2n HCVC aadesh 2 d8sh CVC 0C 0C andaaz 1n d2z VC CVC ahimsa 1 h3n s2 V CVC CV Text to Speech Architecture ! Phonetic Synthesizer Input Text in UTF8 Encodings Speech Synthesizer Text parser CV Pair Algorithm Text to Phonetic Script Conversion Sound concatenation Grapheme To Phoneme Rules Sound Database Speech Sound Database ! Sound files are gsm compressed i.e. “.gsm” fromat Sound units stored in the database are: CV pairs VC pairs V C Halfs :--:--:--:--:--- 1..33 * 2 4 6 8 9 10 12 13 14 15 2 4 6 8 9 10 12 13 14 15 * 1..34 1..14 1..34 ky kr kl kll kv ksh khy khr khl khv gy gr gl gv gn ghy ghr ghv ghn chy chr chv jy jv ty tr tv thy thr dy dr dv dhy dhr dhv ny nr nv tty ttr ttv ddy ddr ddv py pr pl pll fr fl by br bl bhy bhr bhl my mr vy vr vl Total size of db --- 1 MB Sound Concatenation CV files named consonant number named V files x.y.gsm vowel number x.gsm CV files named vowel number Halfs files vowel number 0C files named consonant number x.y.gsm consonant number named x.y.gsm 2 consonants x.gsm 4 more Files cvoffsets vcoffsets voffsets hoffsets Text to Speech Architecture ! Phonetic Synthesizer Text parser Input Text in UTF8 Encodings Text to Phonetic Script Conversion Speech Synthesizer CV Pair Algorithm Sound concatenation Grapheme To Phoneme Rules Sound Database Speech Extended modules:- S2T T2S Constraints :Future scope :- T2T File Reader S2T Reporter Training is tedious : 2 input Languages. Phone generation of all Indian languages difficult. Can be trained for all Indian languages Increase accuracy Better quality of the text to speech synthesizer modules A larger dictionary approx. 2000-3000 words BOL INDIA BOL PRIVATE LIMITED Masters of Computer Application. Sardar Patel Institute Of Technology. Andheri (West) Mumbai-58 Anasree Chatterjee (Director) Diwa Arunashree (Director) Prof. K.T.Talele (Joint Director) Shivani Nadkarni (Joint Director) Aditya Naravane (Joint Director) “Language to Language Translator – A way To Homogeneous India ” Languator -- especially designed for the 3Ts’ that is Travelers, Tourists and at pars the people who are victims of Transferable jobs. It will also serve to certain extent the needs of S2T Reporters.