A framework for bangla text to speech synthesis
2 likes1,218 views
This document presents a framework for Bangla text-to-speech synthesis. It discusses factors that are important for Bangla speech synthesis like diphones and pronunciation variations based on position. The proposed framework includes text normalization rules, a syllable parser to break text into syllables based on rules, and the selection and normalization of audio files for syllables and diphones. An experiment was conducted with test articles to evaluate the accuracy of synthesized speech, showing some limitations but overall effectiveness of the framework. The framework aims to enable Bangla text-to-speech with a minimal set of audio files.
A framework for bangla text to speech synthesis
- 1. A Framework for Bangla Text to Speech Synthesis Authors K. M. Azharul Hasan, Muhammad Hozaifa, Sanjoy Dutta, Rafsan Zani Rabbi Presented By Sanjoy Dutta Department of Computer Science & Engineering Khulna University of Engineering and Technology, Khulna, Bangladesh. Authors
- 2. Contents • Problem Statement • Factors for Speech Synthesis in Bangla • Proposed Framework • Rules and Structure Development • Syllable Parser Development • Audio File Selection and Normalization • Experimental Analysis & Results • Conclusion 2
- 3. Problem Statement •Develop a framework for Bangla Text to Speech Synthesis. 3
- 4. Contents • Problem Statement • Factors for Speech Synthesis in Bangla • Proposed Framework • Rules and Structure Development • Syllable Parser Development • Audio File Selection and Normalization • Experimental Analysis & Results • Conclusion 4
- 5. Factors for Speech Synthesis in Bangla • Sequential flow of diphones A diphone is a set of two adjacent phonemes where the transition between two phonemes are modelled, usually from the middle of the first phoneme to the middle of the second phoneme. A phoneme is a sound or a group of different sounds perceived to have the same function by speakers of the language or dialect in question. Like in English for K/C phoneme: Skill, School. • Position vs. Pronunciation Three kinds of position occurs of consonant and vowels: Constant Vowel(CV) Vowel Constant(VC) Vowel Constant Vowel(VCV) 5
- 6. Contents • Problem Statement • Factors for Speech Synthesis in Bangla • Proposed Framework • Rules and Structure Development • Syllable Parser Development • Audio File Selection and Normalization • Experimental Analysis & Results • Conclusion 6
- 7. Proposed Framework Structure and Rules • Text Normalization: Transforming text into a single standard form. Used when converting text to speech, numbers, dates, acronyms, and abbreviations. Text Normalization for Position vs. Pronunciation. 7
- 8. Normalization rules for ‘ ’ 8
- 9. Normalization rules for ‘ - - - ’ 9
- 11. Syllable Parser In Action 11
- 12. Contents • Problem Statement • Factors for Speech Synthesis in Bangla • Proposed Framework • Rules and Structure Development • Syllable Parser Development • Audio File Selection and Normalization • Experimental Analysis & Results • Conclusion 12
- 13. Audio File Selection and Normalization Total 39 consonants 11 vowels in Bangla After Reduction 28 independent consonants 8 (the vowel ’ ‘ is the exception) vowel 13
- 14. Audio File Selection and Normalization Finally 224 (28*8) audio files for the syllables. 28 consonant against 5 vowels to generate 140 (28*5) diphones. In summary, we need (9 vowels, 28 consonants, 224 syllables and 140 diphones) 401 audio files to be created. 14
- 15. Contents • Problem Statement • Factors for Speech Synthesis in Bangla • Proposed Framework • Rules and Structure Development • Syllable Parser Development • Audio File Selection and Normalization • Experimental Analysis & Results • Conclusion 15
- 16. Experimental Analysis and Results Strategy of Analysis: Sample Input Test: Various News Articles from News Portals Listeners Selection: Anonymous Personals Chosen Randomly Accuracy Analysis: Accuracy = 𝑊𝑜𝑟𝑑𝑠 𝑙𝑖𝑠𝑡𝑒𝑛𝑒𝑟𝑠 𝑤𝑒𝑟𝑒 𝑎𝑏𝑙𝑒 𝑡𝑜 ℎ𝑒𝑎𝑟 𝑜𝑛 1𝑠𝑡 𝑎𝑡𝑡𝑒𝑚𝑝𝑡 𝑐𝑙𝑒𝑎𝑟𝑙𝑦∗100 𝑇𝑜𝑡𝑎𝑙 𝑁𝑜. 𝑜𝑓 𝑤𝑜𝑟𝑑𝑠 𝑖𝑛 𝑒𝑣𝑒𝑟𝑦 𝑠𝑎𝑚𝑝𝑙𝑒 16
- 17. Experiment Result Listening Factors: • Duration Synchronization and Merging • Numerical Value like years Constrains in Sample 1: , , , , , , Constrains in Sample 2: , , , , , , 17
- 18. Limitations and Future Works Detect Noun and Adjective words namely ( ) Noun and ( ) Adjective both words should follow the rule 3(a) . But they don't follow the rule 3(a) and their pronunciation is different. 18
- 19. CONCLUSION We believe the proposed framework can be useful for Bangla TTS development to detect the Bangla words with minimum audio file requirement. 19
- 20. Thank You !!! 20