Pith. sign in

REVIEW 4 major objections 6 minor 28 references

RASMALAI: Resources for Adaptive Speech Modeling in Indian Languages with Accents and Intonations

T0 review · 4 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read RASMALAI pairs 13,000 hours of Indian-language speech with 24 million text descriptions, and the resulting TTS approaches human naturalness on seen voices.

desk verdict Useful multilingual TTS dataset with honest human eval, but the headline instruction-following number is circular and uninterpretable as reported. read the letter →

arxiv 2505.18609 v2 pith:LZC43PAM submitted 2025-05-24 cs.CL

Ashwin Sankar , Yoach Lacombe , Sherry Thomas , Praveen Srinivasa Varadhan , Sanchit Gandhi , Mitesh M Khapra This is my paper · ORCID
classification cs.CL
keywords text-promptedtext-to-speechexpressivespeechsynthesisIndianlanguagesmultilingualTTSdescriptiondatasetinstructionfollowingemotionzero-shotstyletransfer
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper claims that controllable, expressive text-to-speech for Indian languages is now possible at scale, and it backs the claim with two artifacts: RASMALAI, a 13,000-hour, 24-million-description corpus across 23 Indian languages plus English, and IndicParlerTTS, the first open-source text-description-guided TTS for these languages. The model is reported to produce near-human naturalness on seen speakers, with a MUSHRA score of 81.7 versus 89.7 for human speech, and to follow text instructions with high fidelity, reaching an IF-BLEU score of 93.18. It also transfers expressive style to speakers and languages it never saw with expressive training data. If true, this lets a user control voice, emotion, accent, and environment by typing a description instead of supplying a reference audio clip.

What carries the argument

The central mechanism is an attribute-extraction and LLM-captioning pipeline: acoustic features such as mean and standard deviation of pitch, C50, SNR, speaking rate, and PESQ are combined with available metadata like age, gender, and speaker identity, plus Ekman emotion labels, then binned into structured tags. A LLaMA-3.1-8B-Instruct model turns each tag set into three caption variants, Descriptive, Concise, and Attribute-Robust, and IndicTrans2 translates the Descriptive captions into the target native languages. These captions convert existing {audio, transcript} pairs into {audio, transcript, description} training triplets, which is what lets a TTS model condition generation on text rather than on a reference audio sample. The model itself is Parler-TTS mini v1 adapted with an expanded Llama2 tokenizer for better subword segmentation in Indian languages.

What would settle it

Run a blind listening test in which human raters match synthesized clips to text descriptions, and compare those judgments with IF-BLEU and the attribute-accuracy scores; if the automatic scores do not track listener judgments for attributes like 'slightly fast pace' or 'great speech quality,' the instruction-following claim is not supported.

Watch

Extended reading notes

Core claim

The central discovery this paper argues for is that an automated pipeline can convert existing speech corpora into a richly described TTS training set large enough to make text-prompted synthesis work for Indian languages. RASMALAI aggregates read, extempore, conversational, studio-recorded, and accented English speech; extracts metadata, emotion labels, and acoustic attributes; discretizes them into bins; and asks an LLM to write three kinds of captions per utterance, with native-language translations for the descriptive captions. Trained on this corpus, IndicParlerTTS is reported to be the first open-source text-description-guided TTS covering 24 languages, with near-human naturalness on seen speakers and strong instruction adherence. The paper also reports robust zero-shot expressive transfer, both to unseen speakers within a language and across related and distant languages, which is what makes the contribution a step toward controllable multilingual synthesis rather than a single-voice demo.

Load-bearing premise

The claim that the model faithfully follows instructions rests on the assumption that re-annotating synthesized audio with the same acoustic feature pipeline used to build the captions is a valid measure of whether the descriptions were obeyed; if that pipeline does not track what listeners actually perceive, the high instruction-following scores could overstate true adherence.

Editorial extensions

If this is right

  • A user can generate speech in any of 24 languages by typing a description of the speaker's voice, accent, emotion, speaking rate, and recording environment, with no reference audio clip required.
  • On the Rasa-13 subjective test, the model scores 81.7 versus 89.7 for human speech, and in Bodo, Maithili, Marathi, and Telugu the gap is within a few points of human naturalness.
  • Automatic evaluation reports an IF-BLEU score of 93.18 for instruction adherence, with attribute-level accuracy between 80.44% and 99.26% across C50, F0 mean, F0 standard deviation, SNR, speaking rate, and PESQ.
  • The model renders emotions for speakers with no expressive training data, including cross-lingually, with MUSHRA scores of 86.73 for native speakers, 80.86 for speakers of related languages, and 76.01 for speakers of distant languages.
  • The released corpus of 13,000 hours and 24 million annotations provides a public training resource for controllable and expressive Indian-language TTS that previously did not exist.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Because the captions are generated from automatically extracted acoustic tags rather than human listening, the dataset encodes machine-measurable attributes; a human annotation study could reveal which descriptions listeners actually perceive and would likely strengthen the instruction-following evaluation.
  • The cross-lingual expressive transfer results suggest the model builds language-independent representations of emotion and style, which could be exploited for tasks beyond TTS, such as style-consistent dubbing, voice editing, or expressive data augmentation for speech recognition.
  • The relatively lower PESQ accuracy hints that studio-only finetuning limits the model's ability to reproduce deliberately degraded or noisy recordings described in the dataset; adding matched noisy finetuning data could test and close that gap.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 6 minor

Summary. The paper introduces RASMALAI, a large-scale speech dataset of roughly 13,000 hours across 24 languages, with 24 million automatically generated text descriptions built from acoustic attributes (pitch, C50, SNR, speaking rate, PESQ) and metadata (age, gender, emotion, style, accent). Using this dataset, the authors train IndicParlerTTS, a multilingual text-prompted TTS system. They evaluate it with MUSHRA naturalness tests on seen speakers in the Rasa-13 benchmark, automatic metrics (CER, WER, MOS, S-SIM, IF-BLEU), attribute-level accuracy, a human emotion classification test, and zero-shot expressive synthesis across native, proximal, and distal speaker groups. The paper's central claims are that IndicParlerTTS achieves near-human naturalness on seen speakers, reliably follows text descriptions, accurately synthesizes specified attributes, and transfers expressive characteristics within and across languages.

Significance. If validated, this would be a substantial contribution: RASMALAI is much larger and more language-diverse than existing text-prompted TTS datasets for Indic languages, and the public release of the dataset and model would fill a clear gap. The human MUSHRA results (Table 3) and the human emotion classification experiment (Figure 2) are genuine strengths and provide credible evidence for naturalness and emotional expressiveness, independently of the pipeline-based evaluations. The attribute coverage comparison in Table 2 is useful. However, the instruction-following claim, which is central to the paper's 'controllable TTS' contribution, currently rests on an evaluation that reuses the same acoustic-attribute extraction and binning pipeline that generated the training descriptions. This makes the headline IF-BLEU of 93.18 an internal-consistency result rather than evidence of human-perceived prompt adherence. The contribution would be strong once this is addressed with a human attribute-perception test or a careful reframing of the claim.

major comments (4)
  1. [Section 3, Tables 5 and 6] The instruction-adherence evaluation is circular. Section 2.2 constructs training descriptions by extracting acoustic attributes, binning them, and asking an LLM to verbalize them, while Section 3 evaluates adherence by 're-annotating acoustic features of the synthesised samples using our pipeline up to the binning stage' and comparing the binned values against the original attributes. Consequently, IF-BLEU (93.18) and attribute accuracy (80-99%) primarily measure the model's ability to invert the same deterministic feature extractor and binner used to create the training labels, not whether a listener perceives 'slightly fast pace' or 'great speech quality' as described. To support the Section 4.2 claim that the model 'reliably follows text descriptions,' a human attribute-perception test is required, in which listeners match or rate synthesized samples against the natural-language prompts; without this, the headline instruction-following claim should be reframed as acoustic-attribute reconstruction accuracy.
  2. [Section 3, IF-BLEU definition] The definition of IF-BLEU is ambiguous as written. The text says the metric is computed by 'constructing prompts from binned values as comma-separated sequences and compar[ing] them to the original instruction from which this sample was synthesised.' If 'the original instruction' is the LLM-generated natural-language sentence from Section 2.2, BLEU between comma-separated attribute tokens and natural-language tokens would be near zero under standard tokenization; if the comparison is instead to the original binned attributes, the metric is an attribute-reconstruction score, not the instruction-following BLEU introduced in LibriTTS-P. Please specify exactly what the reference string is, how tokenization and punctuation are handled, and whether the score is averaged over the three prompt types or only over one of them. This is necessary for interpreting the reported value of 93.18.
  3. [Section 4.1 and Section 4.4] The evaluation scope is narrower than the claims. The main naturalness and instruction-following results (Tables 3-6) are reported only for the Rasa-13 test set, despite the system covering 24 languages. Table 7 reports zero-shot expressive MUSHRA scores for Native, Proximal, and Distal groups, but does not state the number of speakers, languages, or utterances per group, nor does it include a baseline comparator. In addition, the only comparative baseline, IndicVC, is a zero-shot speech-prompted TTS, not a text-description-guided system, so the comparison in Tables 4 cannot benchmark instruction-following capability. Please either extend the evaluation to additional languages, add an appropriate text-prompted baseline, or temper the 'multilingual' and 'new standard' claims to match the evidence.
  4. [Section 3, Table 5] The automatic metrics in Table 5 (CER, WER, MOS, S-SIM, IF-BLEU) are reported as point estimates without confidence intervals or significance tests, and the size of the evaluation subset is not specified. Given that Section 4.1 uses language such as 'significantly surpasses' for the MUSHRA comparisons, the same standard should be applied to the automatic metrics. Please provide confidence intervals or significance tests for these values, and state how many utterances were used in the TTS subset of the test set.
minor comments (6)
  1. [Section 3] There is a typo in the evaluation paragraph: 'by by re-annotating' should read 'by re-annotating.'
  2. [Section 4.1] The text contains the erroneous token 'LERTTS' where 'IndicParlerTTS' is evidently intended.
  3. [Table 1] The example in Table 1 is internally inconsistent: the attribute list says 'Slightly close sounding,' but the descriptive caption says 'slightly roomy environment.' These are opposite descriptions; please correct the example so the caption matches the attribute list.
  4. [Section 2.2] PESQ is a full-reference metric, but the paper does not state how it is computed for utterances that lack a reference signal. Please clarify the estimation procedure used in the attribute extraction pipeline.
  5. [Section 6] The abstract says the authors 'will release' the dataset and code, while the conclusion says 'We release RASMALAI, along with our models and code.' Please make the availability status consistent and state the intended release license and platform.
  6. [Table 6] The caption for Table 6 refers to 'the TTS subset of the test set,' but this subset is not defined anywhere. Please define it and report the number of utterances used for the attribute-accuracy evaluation.

Circularity Check

1 steps flagged · score 6.0 of 10

Instruction-following metrics (IF-BLEU 93.18 and Table 6) are computed with the same acoustic extraction-and-binning pipeline that created the training descriptions, so they certify internal label consistency rather than independently verified perceived prompt adherence.

  1. self definitional [Section 3 (Evaluation, Tables 5-6); grounded in the description-generation pipeline of Section 2.2]
    "We evaluate instruction adherence by by re-annotating acoustic features of the synthesised samples using our pipeline up to the binning stage. We then compute IF-BLEU [7] by constructing prompts from binned values as comma-separated sequences and compare them to the original instruction from which this sample was synthesised. We also report attribute-level accuracy in Table 6, which is calculated directly by comparing the binned values against the original attributes used to create the description."

    Section 2.2 creates each prompt by extracting acoustic features (pitch, C50, SNR, speaking rate, PESQ), binning them, and asking Llama-3.1-8B-Instruct to verbalize the bins. The evaluation applies the same extraction-and-binning function to synthesized audio and scores agreement with those same bins. Thus IF-BLEU 93.18 and the 80-99% attribute accuracies measure how well the output re-maps onto the paper's own annotation bins (audio -> bins -> LLM prompt -> model -> audio -> bins -> compare), not whether listeners perceive 'slightly fast pace', 'slightly roomy environment', or 'great speech quality'.

full rationale

The human-rated results (MUSHRA naturalness, emotion classification, speaker similarity) are externally grounded and are not circular. The same is true of CER/WER/MOS. However, the paper's central capability claim, 'reliably follow text descriptions' and 'accurately synthesize specified attributes', rests on IF-BLEU (93.18) and attribute accuracy (Table 6), both computed by re-annotating synthesized audio with the same acoustic extraction and binning pipeline that produced the training descriptions in Section 2.2. This makes the instruction-following evaluation a closed loop: the prompt is a verbalization of bins, and the score is agreement with those same bins after synthesis. It is therefore a renamed bin-reconstruction metric rather than an independent test of instruction adherence. No self-citation chain or uniqueness theorem is load-bearing here, and the external human evaluations keep the circularity partial rather than total. Score 6.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The paper's contributions rest on several upstream data quality and tooling assumptions, none of which are validated with human checks. The most consequential choice is the discretization of acoustic attributes, which appears both in annotation generation and in evaluation, creating a potential self-consistency loop.

free parameters (3)
  • Acoustic attribute discretization bins (SNR, C50, F0 mean/std, speaking rate, PESQ) = Not disclosed; follows/extended from Parler-TTS pipeline [8]
    These bins determine the vocabulary of descriptions and the evaluation bins; choices affect both training annotations and IF-BLEU/attribute accuracy scores.
  • LLM prompt templates for description generation = Not provided in paper
    The exact prompts given to Llama-3.1-8B-Instruct shape description diversity and style; without them, reproduction is incomplete.
  • Finetune set selection criteria (18 languages, 1,804 hours) = Not specified
    Rule for selecting the studio-quality finetune subset is not stated, affecting speaker consistency and quality results.
assumptions (4)
  • domain assumption Aggregated source datasets (IndicVoices, RASA, etc.) contain accurate transcripts and reliable speaker/emotion metadata.
    The quality of RASMALAI depends on the correctness of transcripts and labels from external corpora; errors would propagate to descriptions and model training.
  • domain assumption Acoustic features (pitch, C50, SNR, PESQ) estimated by the cited toolkits are accurate across the diverse recording conditions of the source datasets.
    Descriptions are generated from these features; if estimates are noisy, the text annotations and evaluation bins are compromised.
  • domain assumption Llama-3.1-8B-Instruct and IndicTrans2 produce faithful, fluent descriptions and translations at scale.
    No human validation of description quality is reported; the entire annotation pipeline relies on unverified LLM and MT outputs.
  • domain assumption Parler-TTS architecture and training recipe transfer to 24 languages with the expanded tokenizer.
    The model is built from Parler-TTS mini v1 without architectural changes beyond tokenizer expansion; no ablation on language count is provided.

how reviews work

0 comments
Cite this review

Pith. "Pith review of RASMALAI: Resources for Adaptive Speech Modeling in Indian Languages with Accents and Intonations." pith.science (2026). https://pith.science/paper/LZC43PAM

@misc{pith2026250518609,
  author       = {Pith},
  title        = {Pith review of: RASMALAI: Resources for Adaptive Speech Modeling in Indian Languages with Accents and Intonations},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/LZC43PAM}},
  note         = {Machine review of arXiv:2505.18609}
}
read the original abstract

We introduce RASMALAI, a large-scale speech dataset with rich text descriptions, designed to advance controllable and expressive text-to-speech (TTS) synthesis for 23 Indian languages and English. It comprises 13,000 hours of speech and 24 million text-description annotations with fine-grained attributes like speaker identity, accent, emotion, style, and background conditions. Using RASMALAI, we develop IndicParlerTTS, the first open-source, text-description-guided TTS for Indian languages. Systematic evaluation demonstrates its ability to generate high-quality speech for named speakers, reliably follow text descriptions and accurately synthesize specified attributes. Additionally, it effectively transfers expressive characteristics both within and across languages. IndicParlerTTS consistently achieves strong performance across these evaluations, setting a new standard for controllable multilingual expressive speech synthesis in Indian languages.

Figures

Figures reproduced from arXiv: 2505.18609 by the authors.

Figure 2
Figure 2. Confusion plot for Perceptual Emotion Classification [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

28 extracted references · 19 canonical work pages

  1. [1]

    RASMALAI: Resources for Adaptive Speech Modeling in Indian Languages with Accents and Intonations

    Introduction Text-to-Speech (TTS) synthesis has evolved beyond generating speech from a fixed set of voices using high-quality record- ings, with recent advancements exploring speech- and text- prompted TTS systems. Speech-prompted TTS [1, 2, 3, 4] synthesizes speech by replicating the style and speaker char- acteristics from a provided audio prompt. In c...

    work page Pith review arXiv 2025

  2. [2]

    R ASMALAI : An Annotated Corpus for Controllable Multilingual TTS Below we describe (i) existing TTS datasets from which{audio, text} pairs were collated (ii) our approach for generating text descriptions for these audios and (iii) statistics of our dataset. 2.1. Collating existing TTS datasets We first collate{audio, text} pairs from multiple existing In...

    work page

  3. [3]

    Experimental Setup Model: Building on Parler-TTS mini v1, we develop I NDIC - PARLER TTS to support Indian languages by replacing the de- fault tokenizer in Parler-TTS with an expanded Llama2 tok- Figure 1: Comparison of durations for RASMALAI -PRETRAIN , RASMALAI -FINETUNE and INDIC VOICES -R across 24 languages. asm ben brx hne doi eng nep ori pan san s...

    work page

  4. [4]

    Results We evaluate I NDIC PARLER TTS on naturalness, style adher- ence, expressivity, and multilinguality. 4.1. Approaching Human-level Synthesis on Seen Speakers Table 3 highlights languages where I NDIC PARLER TTS ap- proaches human-level naturalness, while Table 4 reports cases where our model significantly outperforms the previous best- performing sy...

    work page

  5. [5]

    However, most efforts are limited to English [5, 7], and cover only a narrow set of attributes with restricted domain diversity

    Related Work Resources and Models for Text-Prompted TTS: Recent works [5, 6, 8, 7] have explored large-scale dataset annota- tion to improve TTS control through diverse speech attributes. However, most efforts are limited to English [5, 7], and cover only a narrow set of attributes with restricted domain diversity. Among these, only [7] publicly releases ...

    work page

  6. [6]

    Using this dataset, we train INDIC PARLER TTS, the first multilingual text- prompted TTS system for Indic languages

    Conclusion We present RASMALAI , a dataset with rich textual descriptions for 13,000 hours of speech across 24 languages, covering di- verse speakers, emotions, and styles, built using structured at- tribute extraction and LLM-based text generation. Using this dataset, we train INDIC PARLER TTS, the first multilingual text- prompted TTS system for Indic l...

    work page

  7. [7]

    V oicecraft: Zero-shot speech editing and text-to-speech in the wild,

    P. Peng, P. Huang, S. Li, A. Mohamed, and D. Harwath, “V oicecraft: Zero-shot speech editing and text-to-speech in the wild,” in Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), ACL 2024, Bangkok, Thailand, August 11-16, 2024 , L. Ku, A. Martins, and V . Srikumar, Eds. Association for Computa...

    doi:10.18653/v1/2024.acl-long.673 2024

  8. [8]

    StyleTTS 2: Towards human-level text-to-speech through style diffusion and adversarial training with large speech language models,

    Y . A. Li, C. Han, V . S. Raghavan, G. Mischler, and N. Mesgarani, “StyleTTS 2: Towards human-level text-to-speech through style diffusion and adversarial training with large speech language models,” in Thirty-seventh Conference on Neural Information Processing Systems , 2023. [Online]. Available: https://openreview.net/forum?id=m0RbqrUM26

    work page 2023

Show all 28 references
  1. [9]

    Naturalspeech 3: Zero-shot speech synthesis with factorized codec and diffusion models,

    Z. Ju, Y . Wang, K. Shen, X. Tan, D. Xin, D. Yang, E. Liu, Y . Leng, K. Song, S. Tang, Z. Wu, T. Qin, X. Li, W. Ye, S. Zhang, J. Bian, L. He, J. Li, and sheng zhao, “Naturalspeech 3: Zero-shot speech synthesis with factorized codec and diffusion models,” in Forty-first Interna...

    2024

  2. [10]

    E2 TTS: embarrassingly easy fully non-autoregressive zero-shot TTS,

    S. E. Eskimez, X. Wang, M. Thakker, C. Li, C. Tsai, Z. Xiao, H. Yang, Z. Zhu, M. Tang, X. Tan, Y . Liu, S. Zhao, and N. Kanda, “E2 TTS: embarrassingly easy fully non-autoregressive zero-shot TTS,” CoRR, vol. abs/2406.18009, 2024. [Online]. Available: https://doi.org/10.48550/a...

    2024 doi arXiv

  3. [11]

    PromptTTS 2: Describing and generating voices with text prompt,

    Y . Leng, Z. Guo, K. Shen, Z. Ju, X. Tan, E. Liu, Y . Liu, D. Yang, leying zhang, K. Song, L. He, X. Li, sheng zhao, T. Qin, and J. Bian, “PromptTTS 2: Describing and generating voices with text prompt,” in The Twelfth International Conference on Learning Representations , 202...

    2024

  4. [12]

    Audiobox: Unified audio generation with natural language prompts,

    A. Vyas, B. Shi, M. Le, A. Tjandra, Y . Wu, B. Guo, J. Zhang, X. Zhang, R. Adkins, W. Ngan, J. Wang, I. Cruz, B. Akula, A. Akinyemi, B. Ellis, R. Moritz, Y . Yungster, A. Rakotoarison, L. Tan, C. Summers, C. Wood, J. Lane, M. Williamson, and W. Hsu, “Audiobox: Unified audio ge...

    2023 doi arXiv

  5. [13]

    Libritts-p: A corpus with speaking style and speaker identity prompts for text-to-speech and style captioning,

    M. Kawamura, R. Yamamoto, Y . Shirahata, T. Hasumi, and K. Tachibana, “Libritts-p: A corpus with speaking style and speaker identity prompts for text-to-speech and style captioning,” in Proc. Interspeech 2024, Sep. 2024

    2024

  6. [14]

    Natural language guidance of high- fidelity text-to-speech with synthetic annotations,

    D. Lyth and S. King, “Natural language guidance of high- fidelity text-to-speech with synthetic annotations,” CoRR, vol. abs/2402.01912, 2024. [Online]. Available: https://doi.org/10. 48550/arXiv.2402.01912

    2024 doi arXiv

  7. [15]

    Mega-TTS 2: Boosting prompting mechanisms for zero-shot speech synthesis,

    Z. Jiang, J. Liu, Y . Ren, J. He, Z. Ye, S. Ji, Q. Yang, C. Zhang, P. Wei, C. Wang, X. Yin, Z. MA, and Z. Zhao, “Mega-TTS 2: Boosting prompting mechanisms for zero-shot speech synthesis,” in The Twelfth International Conference on Learning Representations , 2024. [Online]. Ava...

    2024

  8. [16]

    Re- sources for indian languages,

    A. Baby, A. L. Thomas, N. Nishanthi, T. Consortium et al., “Re- sources for indian languages,” in Proceedings of Text, Speech and Dialogue, 2016

    2016

  9. [17]

    Rasa: Building expressive speech synthesis systems for indian languages in low-resource settings,

    P. Srinivasa Varadhan, A. Sankar, G. Raju, and M. M. Khapra, “Rasa: Building expressive speech synthesis systems for indian languages in low-resource settings,” in Interspeech 2024, 2024, pp. 1830–1834

    2024

  10. [18]

    Limmits’24: Multi- speaker, multi-lingual indic tts with voice cloning,

    A. Singh, A. Nagireddi, D. G, J. Bandekar, R. R, S. Badiger, S. Udupa, P. K. Ghosh, H. A. Murthy, P. Kumar, K. Tokuda, M. Hasegawa-Johnson, and P. Olbrich, “Limmits’24: Multi- speaker, multi-lingual indic tts with voice cloning,” in2024 IEEE International Conference on Acousti...

    2024

  11. [19]

    Indicvoices-r: Unlocking a massive multilingual multi-speaker speech corpus for scaling indian TTS,

    A. Sankar, S. Anand, P. S. Varadhan, S. Thomas, M. Singal, S. Kumar, D. Mehendale, A. Krishana, G. Raju, and M. M. Khapra, “Indicvoices-r: Unlocking a massive multilingual multi-speaker speech corpus for scaling indian TTS,” in The Thirty-eight Conference on Neural Information...

    2024

  12. [20]

    Cross- domain neural pitch and periodicity estimation,

    M. Morrison, C. Hsieh, N. Pruyne, and B. Pardo, “Cross- domain neural pitch and periodicity estimation,” inarXiv preprint arXiv:2301.12258, 2023

    2023 arXiv

  13. [21]

    Brouhaha: multi-task training for voice activity detection, speech-to-noise ratio, and C50 room acoustics estimation,

    M. Lavechin, M. M ´etais, H. Titeux, A. Boissonnet, J. Copet, M. Rivi`ere, E. Bergelson, A. Cristia, E. Dupoux, and H. Bredin, “Brouhaha: multi-task training for voice activity detection, speech-to-noise ratio, and C50 room acoustics estimation,”ASRU, 2023

    2023

  14. [22]

    pyannote.audio: neural building blocks for speaker diarization,

    H. Bredin, R. Yin, J. M. Coria, G. Gelly, P. Korshunov, M. Lavechin, D. Fustes, H. Titeux, W. Bouaziz, and M.-P. Gill, “pyannote.audio: neural building blocks for speaker diarization,” in ICASSP 2020, IEEE International Conference on Acoustics, Speech, and Signal Processing, B...

    2020

  15. [23]

    Indictrans2: Towards high-quality and accessible machine translation models for all 22 scheduled indian languages,

    J. P. Gala, P. A. Chitale, R. AK, V . Gumma, S. Doddapaneni, A. K. M., J. A. Nawale, A. Sujatha, R. Puduppully, V . Raghavan, P. Kumar, M. M. Khapra, R. Dabre, and A. Kunchukuttan, “Indictrans2: Towards high-quality and accessible machine translation models for all 22 schedule...

    2023

  16. [24]

    IndicVoices: Towards building an inclusive multilingual speech dataset for Indian languages,

    T. Javed, J. Nawale, E. George, S. Joshi, K. Bhogale, D. Mehendale, I. Sethi, A. Ananthanarayanan, H. Faquih, P. Palit, S. Ravishankar, S. Sukumaran, T. Panchagnula, S. Murali, K. Gandhi, A. R, M. M, C. Vaijayanthi, K. Karunganni, P. Kumar, and M. Khapra, “IndicVoices: Towards...

    2024

  17. [25]

    Globe: A high-quality english corpus with global accents for zero-shot speaker adaptive text-to- speech,

    W. Wang, Y . Song, and S. Jha, “Globe: A high-quality english corpus with global accents for zero-shot speaker adaptive text-to- speech,” in Interspeech 2024, 2024, pp. 1365–1369

    2024

  18. [26]

    An empirical comparison of vocabulary expansion and initialization approaches for language models,

    N. Mundra, A. N. K. Khandavally, R. Dabre, R. Puduppully, A. Kunchukuttan, and M. M. Khapra, “An empirical comparison of vocabulary expansion and initialization approaches for language models,” in Proceedings of the 28th Conference on Computational Natural Language Learning , ...

    2024

  19. [27]

    Stateful conformer with cache-based inference for streaming automatic speech recognition,

    V . Noroozi, S. Majumdar, A. Kumar, J. Balam, and B. Ginsburg, “Stateful conformer with cache-based inference for streaming automatic speech recognition,” in IEEE International Conference on Acoustics, Speech and Signal Processing, ICASSP 2024, Seoul, Republic of Korea, April ...

    2024

  20. [28]

    Speech quality assessment through mos using non-matching references,

    P. Manocha and A. Kumar, “Speech quality assessment through mos using non-matching references,” in Interspeech 2022, 2022, pp. 654–658

    2022

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.