Elamur Research⌕
☰
home / papers / 2022 / Study of Indian English Pronunciation Variabilities relative to Received Pronunciation

Study of Indian English Pronunciation Variabilities relative to Received Pronunciation

Paper · 2022 · Priyanshi Pal, Shelly Jain, Anil Vuppala et al.

Summary

The paper looks at how Indian English (IE) pronunciation differs from Received Pronunciation (RP). Speech recognition and synthesis for Indian users need this information. Earlier linguistic work described these differences only qualitatively, as phonetic rules without frequency figures, and phonetically labelled IE data is rare. The authors use the Indic TIMIT corpus, which has 80 speakers from six regions of India grouped by the language family of their mother tongue. They expand its phonetic annotation to 15,974 transcribed utterances, 13,632 of them newly annotated for this study. The transcriber's self-consistency was checked with Cohen's Kappa and found to be high. Using these transcriptions, the authors do three things:

  • assign quantitative prevalence measures to rules already proposed in the literature;
  • mine new IE-versus-RP rules that earlier work had not described;
  • build a Grapheme-to-Phoneme (G2P) converter from the derived rules and evaluate it with Phoneme Error Rate (PER).

The contribution is a data-grounded, measurable account of IE phonetics in place of anecdotal description, which makes pronunciation modelling for Indian speech systems more practical.

Key points

  • Corpus: Indic TIMIT has 80 L2 English speakers aged 18–60 from IISc Bangalore, about 240 hours in total, with 2,342 TIMIT prompts per speaker.
  • Speaker groups: the six regions map to five native-language groups, mostly Indo-Aryan, Tibeto-Burman and Dravidian languages (e.g., Hindi/Punjabi, Gujarati/Marathi, Kannada/Telugu, Malayalam/Tamil).
  • Transcriptions: 2,342 already existed, made by two linguists. One of those linguists added 13,632 more, giving 15,974 in total, written with 108 IPA symbols.
  • Consistency checks: intra-rater agreement was measured with Cohen's Kappa. Re-transcribing 200 repeated files per batch gave a mean of 0.827. Correcting shuffled transcriptions of the same prompt from different speakers gave 0.84. Values above 0.80 count as strong agreement.
  • Validation: the mined rules are compared with rules from the literature, and a G2P system built on them is scored by PER.

Interesting ideas

  • I1. Existing qualitative rules from the literature are turned into measured ones by counting how often each actually occurs in a sizeable annotated corpus. This separates widespread phenomena from rare ones.
  • I2. Annotator consistency is measured two ways: plain repetition of the same items, and re-correction of shuffled outputs for the same prompt from different sources. The second makes memorised answers less likely.
  • I3. When two sequences differ in length because of insertions or deletions, they are padded with dummy symbols so a chance-corrected agreement score like Kappa can still be computed.
  • I4. Annotation batches are built so that every batch covers all speaker groups. This keeps consistency checks and analyses balanced across varieties.
  • I5. Rules are mined from data and then validated twice: against expert-written rules, and through a downstream system (G2P scored by PER).

For Mova

How these ideas could serve Mova, our small open language model — its files, stages and tests:

  • I1 (tree): UD dialect-conversion and treebank-normalisation rules could each carry corpus frequency counts, so rare or doubtful rules are visible and can be deprioritised.
  • I2 (learn): LLM teacher tiers could be re-queried on duplicated and shuffled items, and their self-agreement used to decide whether to trust their silver labels.
  • I3 (learn): the same trick lets several witness parsers or taggers be compared with Kappa even when their tokenizations diverge, rather than with raw accuracy.
  • I4 (learn): silver-data and evaluation batches could be stratified by genre or source (school tests, fables, word problems), so agreement numbers are not dominated by one text type.
  • I5 (tokens): self-induced morphology or lemma rules could be checked both against existing grammar descriptions and by their effect on tagging and lemma error.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Institutions

🎓International Institute of Information Technology, HyderabadHyderabad · India24 works · 1,719 citations🎓Indian Institute of Science BangaloreBengaluru · India8 works · 1,140 citations

📚 Shelves 7The field of science this work belongs to (OpenAlex fields).Computer Science · 9,086 worksArtificial Intelligence · 8,889 worksArts and Humanities · 9,296 worksLanguage and Linguistics · 9,206 worksPsychology · 798 worksExperimental and Cognitive Psychology · 463 worksSignal Processing · 213 works

🧭 Topics 1Research subjects and languages the work is about.English · 1,373 works

Cite

Cite as Priyanshi Pal, Shelly Jain, Anil Vuppala, Chiranjeevi Yarra and Prasanta Ghosh. 2022. Study of Indian English Pronunciation Variabilities relative to Received Pronunciation. arXiv:2204.06502.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.48550/arxiv.2204.06502 · arXiv 2204.06502

@misc{pal2022study,
  title = {{Study of Indian English Pronunciation Variabilities relative to Received Pronunciation}},
  author = {Priyanshi Pal, Shelly Jain, Anil Vuppala, Chiranjeevi Yarra and Prasanta Ghosh},
  year = {2022},
  doi = {10.48550/arxiv.2204.06502},
  eprint = {2204.06502},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.2204.06502}
}

References

31 references, 1 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. C. Chandramouli and R. General, “Census of india 2011,” Provisional Population Totals. New Delhi: Government of India, pp. 409–413, 2011. find
  2. S. Kishore, R. Kumar, and R. Sangal, “A data driven synthesis approach for indian languages using syllable as basic unit,” in Proceedings of Intl. Conf. on NLP (ICON), 2002, pp. 311–316. find
  3. S. Sitaram, V. Manjunatha, V. Bharadwaj, M. Choudhury, K. Bali, and M. Tjalve, “Discovering canonical indian english accents: a crowdsourcing-based approach,” in Proceedings of the Eleventh International Conference on Language Resources and Evaluation (LREC 2018), 2018. find
  4. S. R. Vignesh, S. A. Shanmugam, and H. A. Murthy, “Significance of pseudo-syllables in building better acoustic models for indian english tts,” in 2016 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2016, pp. 5620–5624. find
  5. R. Saikia and S. R. Singh, “Generating manipuri english pronunciation dictionary using sequence labelling problem,” in 2016 International Conference on Asian Language Processing (IALP), 2016, pp. 67–70. find
  6. R. Kumar, R. Gangadharaiah, S. Rao, K. Prahallad, C. P. Rosé, and A. W. Black, “Building a better indian english voice using” more data”.” in SSW. Citeseer, 2007, pp. 90–94. find
  7. D. Vazhenina and K. Markov, “Phoneme set selection for russian speech recognition,” in 2011 7th International Conference on Natural Language Processing and Knowledge Engineering. IEEE, 2011, pp. 475–478. find
  8. M. C. Anil and S. Shirbahadurkar, “Phoneme selection rules for marathi text to speech synthesis with anuswar places,” in Advances in Signal Processing and Intelligent Recognition Systems. Springer, 2016, pp. 501–509. ↗
  9. X. Huang, X. Jin, Q. Li, and K. Zhang, “On construction of the asr-oriented indian english pronunciation dictionary,” in Proceedings of the 12th Language Resources and Evaluation Conference, 2020, pp. 6593– 6598. find
  10. R. Weide, “The carnegie mellon pronouncing dictionary [cmudict. 0.6],” Version 0.6. Available at [www. speech. cs. cmu. edu/cgi-bin/cmudict], 2005. find
  11. K. Pisegna and V. Volenec, “Phonology and phonetics of l2 telugu english,” Studies in Linguistics and Literature, vol. 5, pp. 46–69, 02 2021. ↗
  12. C. Yarra, R. Aggarwal, A. Rajpal, and P. K. Ghosh, “Indic timit and indic english lexicon: A speech database of indian speakers using timit stimuli and a lexicon from their mispronunciations,” in 2019 22nd Conference of the Oriental COCOSDA International Committee for the Co-ordination and Standardisation of Speech Databases and Assessment Techniques (O-COCOSDA). IEEE, 2019, pp. 1–6. find
  13. V. Zue, S. Seneff, and J. Glass, “Speech database development at mit: Timit and beyond,” Speech communication, vol. 9, no. 4, pp. 351–356, 1990. ↗
  14. R. K. Bansal and J. B. Harrison, Spoken English: A Manual of Speech and Phonetics. Sangam, 1994. find
  15. S. Pingali, Indian English. Edinburgh University Press, 2009. find
  16. H. Sirsa and M. A. Redford, “The effects of native language on indian english sounds and timing patterns,” Journal of phonetics, vol. 41, no. 6, pp. 393–406, 2013. ↗
  17. R. K. Bansal, “The pronunciation of english in india,” Studies in the pronunciation of English: A commemorative volume in honour of AC Gimson, pp. 219–230, 1990. find
  18. R. Mesthrie, “Introduction: varieties of english in africa andsouth and southeast asia,” in 4 Africa, South and Southeast Asia. De Gruyter Mouton, 2008, pp. 23–34. ↗
  19. J. C. Wells, Accents of English: Volume 3: Beyond the British Isles. Cambridge University Press, 1982, vol. 3. find
  20. P. Sailaja, “Indian english: Features and sociolinguistic aspects,” Language and Linguistics Compass, vol. 6, no. 6, pp. 359–370, 2012. ↗
  21. R. Gargesh, “Indian english: Phonology,” in A handbook of varieties of English. De Gruyter Mouton, 2008, pp. 992–1002. ↗
  22. J. Cohen, “A coefficient of agreement for nominal scales,” Educational and psychological measurement, vol. 20, no. 1, pp. 37–46, 1960. ↗
  23. T. Robinson, “Beep dictionary,” BEEP dictionary, 1996. find
  24. T. Robinson, J. Fransen, D. Pye, J. Foote, and S. Renals, “Wsjcamo: a british english speech corpus for large vocabulary continuous speech recognition,” in 1995 International Conference on Acoustics, Speech, and Signal Processing, vol. 1. IEEE, 1995, pp. 81–84. find
  25. V. Chand, “[v] at is going on? local and global ideologies about indian english,” Language in Society, vol. 38, no. 4, pp. 393–419, 2009. ↗
  26. B. B. Kachru, Asian Englishes: beyond the canon. Hong Kong University Press, 2005, vol. 1. find
  27. ——, English in South Asia, ser. The Cambridge History of the English Language. Cambridge University Press, 1994, vol. 5, p. 497–553. find
  28. S. Jiampojamarn, G. Kondrak, and T. Sherif, “Applying many-to-many alignments and hidden markov models to letter-to-phoneme conversion,” in Human Language Technologies 2007: The Conference of the North American Chapter of the Association for Computational Linguistics; Proceedings of the Main Conference. Rochester, New York: Association for Computational Linguistics, April 2007, pp. 372–379. [Online]. Available: http://www.aclweb.org/anthology/N/N07/N07-1047 ↗
  29. E. Ristad and P. Yianilos, “Learning string-edit distance,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 20, no. 5, pp. 522–532, 1998. ↗
  30. D. Aha, D. Kibler, and M. Albert, “Instance-based learning techniques,” Machine Learning, vol. 6, pp. 37–66, 1991. find
  31. M. Bisani and H. Ney, “Joint-sequence models for grapheme-to-phoneme conversion,” Speech Communication, vol. 50, no. 5, pp. 434–451, 2008. [Online]. Available: https://www.sciencedirect.com/science/article/pii/S0167639308000046 This figure "fig1.png" is available in "png" format from: http://arxiv.org/ps/2204.06502v2 ↗

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/pal-2022-study-indian-english-pronunciation-variabilities/