Automatic Speech Recognition in Sanskrit: A New Speech Corpus and Modelling Insights

Abstract

Automatic speech recognition (ASR) in Sanskrit is interesting, owing to the various linguistic peculiarities present in the language. The Sanskrit language is lexically productive, undergoes euphonic assimilation of phones at the word boundaries and exhibits variations in spelling conventions and in pronunciations. In this work, we propose the first large scale study of automatic speech recognition (ASR) in Sanskrit, with an emphasis on the impact of unit selection in Sanskrit ASR. In this work, we release a 78 hour ASR dataset for Sanskrit, which faithfully captures several of the linguistic characteristics expressed by the language. We investigate the role of different acoustic model and language model units in ASR systems for Sanskrit. We also propose a new modelling unit, inspired by the syllable level unit selection, that captures character sequences from one vowel in the word to the next vowel. We also highlight the importance of choosing graphemic representations for Sanskrit and show the impact of this choice on word error rates (WER). Finally, we extend these insights from Sanskrit ASR for building ASR systems in two other Indic languages, Gujarati and Telugu. For both these languages, our experimental results show that the use of phonetic based graphemic representations in ASR results in performance improvements as compared to ASR systems that use native scripts. 1 * Joint first author 1 Dataset and code can be accessed from www.cse.iitb.ac.in/~asr and https://github. com/cyfer0618/Vaksanca.git.

Improving speech recognition systems for the morphologically complex Malayalam language using subword tokens for language modeling

“…Subword token-based language modeling has been proposed for applications in speech recognition [7,11,12,22,23], statistical machine translation [24], neural machine translations [20,21] and handwriting recognition [25]. The choice of subword tokens used in language modeling impacts the performance of the model on many downstream tasks [26] including speech recognition [7].…”

“…Orthographic syllable-based tokenization of text was proposed by Kunchukuttan et al for statistical machine translation applications [24]. Splitting the tokens based on vowels and adjacent consonants, named vowel segmentation, was proposed by Adiga et al and employed in the context of Sanskrit speech recognition [22]. These two methods segment text into syllable-like units at valid pronunciation boundaries.…”

IndicSUPERB: A Speech Processing Universal Performance Benchmark for Indian Languages

“…Most of these datasets such as MUCS (Diwan et al 2021), MSR (Srivastava et al 2018), Gramvaani (Bhanushali et al 2022), Crowdsourced Speech Corpora (CSC) (Kjartansson et al 2018a), IISC-MILE Corpus (Ayyavu, Pilar, and G 2022), Crowdsourced Multispeaker Speech Dataset (CSD) (He et al 2020), Kashmiri Data Corpus (KDC) 3 , Common Voice, IIIT-H Indic Speech Databases (ISD) (Prahallad et al 2012), Hindi-Tamil ASR Challenge 4 (HTC), Vāksañcayah . (VAC) (Adiga et al 2021), IIIT-H Telugu Corpus (Ganesh et al 2021) (ITH), IITB Marathi Corpus (IMC) (Abraham et al 2020a) and SMC Malayalam Corpus 5 contain only ASR data. Further, most of them support very few languages.…”

Mlphon: A Multifunctional Grapheme-Phoneme Conversion Tool Using Finite State Transducers

“…It has been demonstrated in literature that subword based models are better in capturing language features for morphologically complex languages [18]. Syllables serve as a good choice of subwords for practical applications including automatic speech recognition [67] that takes care of OOV scenarios. Orthographic syllable units have proven to be more effective in statistical machine translation, than other basic units (word, morpheme and character) when trained over small parallel corpora [17].…”