Pith. sign in

REVIEW 4 major objections 5 minor 29 references

SraVaani 1.0: Scaling Inclusive Speech Recognition for Indic Languages

T0 review · 4 major / 5 minor · reviewed 2026-08-14 · deepseek-v4-flash

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

Pith's one-line read SraVaani-1.0 can transcribe 65 Indian languages and dialects, including 44 low-resource and tribal languages with no competing system.

desk verdict A useful and honest Indic-ASR resource paper whose 44-language coverage claim is real but measured only in-domain; the alignment stage it bills as a contribution is never ablated. read the letter →

arxiv 2608.08235 v2 pith:J6CYBZC2 submitted 2026-08-08 eess.AS

Sujith Pulikodan , Agneedh Basu , Pavan Kumar J , Pranav D Bhat , Suryansh Shukla , Nihar Desai , Prasanta Kumar Ghosh This is my paper · ORCID
classification eess.AS
keywords multilingualASRIndiclanguageslow-resourcespeechrecognitiontribalself-supervisedlearningaudio-imagealignmentFastConformerhybridTDT-CTC
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

SraVaani-1.0 is a multilingual automatic speech recognition model, trained from scratch on public Indian speech data, that transcribes 65 Indian languages and dialects. The paper's central claim is coverage: it is the only open, evaluated system that produces transcripts for 44 low-resource and tribal languages, from Garo to Nyishi, that no baseline ASR system supports. Against three multilingual baselines across eight benchmarks, it reports the best word error rate on 10 of the 17 comparable languages and the lowest mean WER, 28.4%, on that set, while staying competitive on high-resource languages. The breadth is attributed to a three-stage recipe: self-supervised pretraining on unlabelled VAANI audio, a transcription-free audio–image alignment stage, and multilingual fine-tuning with a hybrid TDT–CTC decoder. The authors' stated significance is that open ASR can now serve communities that previously had no transcription technology at all.

What carries the argument

The argument is carried by a three-stage training pipeline built on the FastConformer encoder, a Conformer variant with 8× depthwise-strided subsampling and 17 Transformer layers. Stage one applies a wav2vec 2.0-style contrastive objective to 31,255 hours of unlabelled VAANI speech. Stage two aligns the audio encoder to a frozen SigLIP2 vision encoder through a sigmoid contrastive loss, using an attention-pooling alignment head and MAXSIM late-interaction similarity over 11.85 million audio–image pairs; this head is discarded afterwards. Stage three attaches a Hybrid Token-and-Duration Transducer with a CTC auxiliary head and fine-tunes on 31,263 hours of labelled speech from 24 public corpora, using a shared 5,000-unit SentencePiece tokenizer. The alignment stage is the distinctive mechanism: it injects semantic signal from images into speech representations without any transcripts, and the paper credits it with improving low-resource recognition.

What would settle it

Collect fresh recordings of speakers of Garo, Mizo, Bhojpuri, and Nyishi from community radio or field interviews, entirely outside the VAANI corpus, transcribe them manually, and compare SraVaani-1.0's WER on these clips with the VAANI-reported figures; if the error rates climb sharply or the language is misidentified, the unique-coverage claim fails.

Watch

Extended reading notes

Core claim

The discovery, stated as the authors would state it, is that a single ASR system trained entirely on public data can cover 65 Indian languages and dialects, including 44 that no released system transcribes, without sacrificing accuracy on the high-resource languages. Those 44 are scored on the VAANI benchmark, the only test set available for them; across the 32 languages with at least 0.1 hours of test audio, the model reports a median WER of 50.65% and a mean of 50.2%, with strong results for languages with high-resource relatives (Garo 9.5%, Mizo 25.3%) and weak results for isolates such as Nyishi (93.9%). Since none of the three baselines claims support for any of these languages, the paper presents SraVaani-1.0 as the first open transcription capability for that set.

Load-bearing premise

The entire coverage claim rests on the VAANI test set being a fair measure of real-world transcription quality for the 44 languages, and the paper itself acknowledges that training and evaluation data come from the same underlying corpus, so an independent test set could lead to different conclusions.

Editorial extensions

If this is right

  • Practitioners can deploy an open checkpoint to transcribe 44 Indian languages and dialects that previously had no ASR option at all.
  • The open weights and public training recipe give future low-resource ASR work a new baseline to beat on VAANI and on newly collected corpora.
  • The transcription-free audio–image alignment stage offers a reusable way to improve low-resource accuracy without paying for more transcriptions, as long as paired images and speech are available.
  • A single shared 5,000-unit tokenizer across 65 languages is reported to be sufficient for competitive results, indicating that script diversity need not require separate vocabularies.

Reading between the lines

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

  • A natural next experiment is an ablation that removes the audio–image stage; the paper does not isolate its contribution, so it is untested whether low-resource gains come from the alignment signal or simply from additional training.
  • If independent benchmarks confirm the VAANI numbers, the picture-prompt collection protocol used for VAANI could become a template for bootstrapping ASR on other undocumented languages, since it yields pretraining audio and alignment supervision together, with no transcripts.
  • The same three-stage recipe should transfer to low-resource language families outside India, provided paired image–speech data and a small transcribed seed exist for them.
  • The 44-language set is a lower bound for the model's practical reach: fine-tuning the open weights on new field recordings could extend transcription to additional dialects.
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 / 5 minor

Summary. The paper describes SraVaani-1.0, a FastConformer-based multilingual ASR system covering 65 Indian languages and dialects. Training proceeds in three stages: wav2vec 2.0-style contrastive self-supervised pretraining on 31,255 h of VAANI audio; a transcription-free audio–image alignment stage using 11.85M VAANI picture–prompt pairs and a frozen SigLIP2 encoder with a sigmoid contrastive loss; and supervised fine-tuning of a Hybrid TDT-CTC decoder on 30,565 h from 24 public datasets. Evaluation compares SraVaani-1.0 with Gemini 3 Flash, Sarvam Saaras v3, and IndicConformer across eight benchmarks. The authors report the lowest mean WER (28.4%) over 17 comparable Indic languages, best WER on 28 of 68 language–dataset pairs, and unique transcription coverage of 44 low-resource/tribal languages evaluated on the VAANI benchmark, with a median WER of 50.65% on 32 reported splits. The paper candidly states in Section 7 that training and evaluation data are derived from the same underlying VAANI corpus.

Significance. If the results hold, the model would be a meaningful step for open ASR coverage of under-resourced Indic languages. The release of model weights and the unusually detailed training configuration are strengths, as is the candid Section 7 limitation statement. The headline coverage claim, however, rests entirely on in-domain VAANI evaluation, and the audio–image alignment stage is not isolated by any ablation. These issues make the significance conditional rather than established; the paper is nevertheless a valuable system report if the claims are re-scoped to the evidence actually presented.

major comments (4)
  1. [§7 / §6.3 / §3.1 / §4.4 / §5.5] The central claim that SraVaani-1.0 provides transcription capability for 44 low-resource and tribal languages is supported only by WERs on the VAANI test set, and Section 7 explicitly concedes that training and evaluation "are derived from the same underlying dataset." The VAANI corpus is the source of the SSL pretraining audio (§3.1), of the 11,848,593 audio–image alignment pairs (§4.4), and, through reference [7] in §5.5, part of the fine-tuning mixture. Section 7 also notes that independent evaluation "could lead to different conclusions." I therefore cannot treat the reported 50.65% median WER as evidence of out-of-domain transcription capability. Please evaluate on independently collected data for at least a subset of the 44 languages, or, absent that, re-word the abstract and contributions from "provides transcription capability" to "achieves these in-domain WERs on the VAANI benchmark."
  2. [§4.2–§4.3 / Table 1] The audio–image alignment stage is claimed to "improve downstream recognition, particularly for low-resource languages," but no ablation isolates its effect. The final model is trained as pretraining → alignment → fine-tuning, and no comparison is reported for the pipeline with the alignment stage removed (SSL → fine-tuning) or with a control using mismatched audio–image pairs. Because the claim is load-bearing for the paper's three-stage contribution, please add these ablations. Without them, the abstract's attribution of the reported coverage and accuracy to multimodal alignment is not established.
  3. [§6.3 / Table 4] The claim that no competing system provides transcription capability for the 44 languages rests on defining capability as official language support. Table 4 itself shows that IndicConformer, when given a script-matched language identifier, produces output for 19 of the 32 listed languages, and Gemini 3 Flash and Sarvam Saaras v3 produce output for all 32 despite being labeled unsupported. If a functional definition of capability is used, the exclusivity claim is not supported; if an official-support definition is used, it needs to be stated explicitly and distinguished from the claim that "no transcription system exists." Please report and compare all systems that produce output for a language rather than leaving columns blank, or clearly separate the official-support and functional-capability statements.
  4. [§6.3 / Table 4] The VAANI-only WER comparisons are statistically fragile: 23 of the 32 reported splits have under 30 minutes of test audio, 12 of the 44 languages are omitted from Table 4 with results only "available in the release artefacts," and no confidence intervals or significance tests are provided. The median WER of 50.65% and the ordering among systems should therefore be treated as descriptive rather than definitive. Please report confidence intervals or per-language utterance counts, include the omitted 12 languages in a supplementary table, and avoid strong comparative claims based on splits containing only a few minutes of audio.
minor comments (5)
  1. [Abstract / §1] There are typographical issues in the abstract and introduction, including "V AANI" spacing, the missing "In" before "the first stage," and the duplicated phrase "Sarvam Saaras v3 covers also covers" in §1.
  2. [Table 3 / §6.2] The dagger marker on the Tamil row for Sarvam Saaras v3 ("35.2 †") is not explained in the text; please clarify whether that cell is excluded from the row comparison and how the best value is determined for a row with an excluded entry.
  3. [Figure 3] The figure caption says "All 49 Indic languages," while the paper claims 65 supported languages and 44 unique VAANI-only languages; please reconcile these counts and state whether the 12 omitted VAANI-only languages appear in the figure.
  4. [§6.2] The phrase "a large number of language-dataset pairs" is vague; please state the exact denominator and how ties were handled when reporting 28 of 68 best pairs.
  5. [§4.4 / §7] Minor typos include "overlaping" in §4.4 and inconsistent capitalization of "Vaani" versus "VAANI" across the manuscript; these should be corrected in revision.

Circularity Check

0 steps flagged · score 2.0 of 10

No construction-level circularity: the pipeline is an empirical recipe with no fitted parameter renamed as a prediction, but the 44-language coverage claim rests on the authors' own VAANI corpus used both for training and evaluation, a limitation the paper itself discloses.

full rationale

Walking the claimed derivation chain, SraVaani-1.0's training pipeline (SSL pretraining on VAANI, audio-image alignment on VAANI pairs, supervised fine-tuning on 24 public corpora) is an empirical recipe, not a derivation in which an output equals an input by construction. The WER results in Tables 3 and 4 are measured, not fitted: no parameter is tuned to a subset and then reported as a prediction of a closely related quantity. External benchmarks (CommonVoice, FLEURS, IndicTTS, Kathbath, RESPIN, GramVaani, MUCS) provide independent evaluation for the 17 comparison languages, so the system's central comparative content is not purely self-referential. The 44 unique-language coverage claim is the only part that depends on VAANI alone, and VAANI is both an author-created corpus and a training source; Section 7 explicitly concedes that training and evaluation 'are derived from the same underlying dataset' and that independent evaluation 'could lead to different conclusions.' That is an external-validity limitation, not a circular reduction: the VAANI test split is disjoint from training and the reported WERs are free variables that could have been poor. The self-citation to [7] is present and mildly load-bearing for the breadth claim, but it is not an unverified theorem invoked to forbid alternatives, and no equation or fitted value is renamed as a prediction. Per the proportionality rule, this warrants a low score rather than a circularity finding.

Assumptions & free parameters 6 free parameters · 5 assumptions · 0 invented entities

The central claim is empirical and rests primarily on data and training choices rather than on a mathematical derivation. The entries below list the hand-chosen hyperparameters and the unstated assumptions about data quality and evaluation that the reader must accept. None of them is an invented physical entity; the model is a conventional neural system.

free parameters (6)
  • CTC loss weight lambda_CTC = 0.3
    Hand-chosen coefficient balancing the TDT and CTC objectives in the final loss (Section 5.1). It affects how much the model relies on each head and is not derived from first principles.
  • BPE vocabulary size = 5,000
    SentencePiece subword vocabulary built from the combined transcriptions (Section 5.2). The size is a design choice that trades coverage across 65 scripts against unit granularity.
  • Contrastive temperature kappa = 0.1
    Temperature of the wav2vec 2.0-style contrastive loss during pretraining (Section 3.2), with 50 distractors. Hand-set hyperparameters that shape the self-supervised objective.
  • Audio and image token counts for alignment = Ka=16, Kv=16
    The audio encoder is pooled to 16 tokens and the top 16 image tokens are selected by L2 norm for the MAXSIM similarity (Section 4.3). These compression choices are chosen by hand.
  • SigLIP temperature and bias = learned
    The alignment loss uses a learnable temperature t and bias b (Section 4.2.1). These are fitted during the alignment stage rather than fixed a priori.
  • SpecAugment masking policy = 2 freq masks width 27, 10 time masks width 5%
    Augmentation parameters applied during pretraining and fine-tuning (Section 2.2). They are chosen by hand and affect generalization.
assumptions (5)
  • domain assumption The VAANI audio-image pairs are semantically related: the prompting image corresponds to the spoken content.
    Section 4.4 assumes the picture-prompt protocol produces meaningful audio-image pairs. If the image is only weakly related to the speech, the alignment stage injects a noisy training signal.
  • domain assumption Frozen SigLIP2-Large embeddings provide useful semantic targets for speech in 105 Indian languages.
    Section 4.3 relies on a vision-language model trained mostly on web data to represent visual concepts across Indic languages. This transfer is assumed, not validated.
  • domain assumption Transcripts and language labels in the 24 public datasets are accurate enough for supervised fine-tuning and evaluation.
    Section 5.5 filters punctuation, digits, and code-switched text but does not report a manual audit of label or transcript quality.
  • domain assumption Disjoint train/test splits of VAANI are sufficient to prevent data leakage and support generalization claims.
    Section 7 acknowledges that training and evaluation come from the same underlying dataset with similar recording conditions, so the split only controls overlap, not distribution shift.
  • domain assumption WER after lowercasing and punctuation removal is a fair cross-lingual metric.
    Section 6.1 applies a single normalization recipe across scripts. Languages with different orthographic conventions may be penalized differently by this recipe.

how reviews work

0 comments
Cite this review

Pith. "Pith review of SraVaani 1.0: Scaling Inclusive Speech Recognition for Indic Languages." pith.science (2026). https://pith.science/paper/J6CYBZC2

@misc{pith2026260808235,
  author       = {Pith},
  title        = {Pith review of: SraVaani 1.0: Scaling Inclusive Speech Recognition for Indic Languages},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/J6CYBZC2}},
  note         = {Machine review of arXiv:2608.08235}
}
read the original abstract

India's linguistic landscape spans over 700 languages and thousands of dialects, yet the vast majority of automatic speech recognition (ASR) systems support only a small fraction of this diversity. We present SraVaani-1.0, a multilingual ASR model covering 65 Indian languages and dialects, many of which currently have no publicly available or competing ASR system. SraVaani-1.0 is built on a FastConformer architecture and trained from scratch through a three stage the first stage, we perform self-supervised pretraining on 31,255 hours of unlabelled speech from the VAANI corpus using a contrastive learning objective. In the second stage, we introduce an audio-image representation alignment stage that leverages the paired images and speech available in the VAANI corpus. This multimodal alignment encourages the speech encoder to learn semantically richer representations by exploiting the relationship between visual context and spoken content, thereby improving downstream recognition, particularly for low resource the final stage, the aligned encoder is fine-tuned end-to-end using a Hybrid Token-and-Duration Transducer (TDT)-CTC decoder on 31,263 hours of labelled multilingual Indian speech compiled from 24 public datasets spanning 65 languages and dialects. We evaluate SraVaani-1.0 against three state-of-the-art multilingual ASR systems across eight benchmarks. SraVaani-1.0 achieves the lowest word error rate (WER) on a large number of language-dataset pairs while remaining competitive with the best-performing systems on high resource importantly, it is the only open-source evaluated model that provides transcription capability for multiple low-resource and tribal Indian languages, which are assessed exclusively on the VAANI benchmark.

Figures

Figures reproduced from arXiv: 2608.08235 by the authors.

Figure 1
Figure 1. Training (left) and validation (right) contrastive loss vs. step. [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Three-stage pipeline: audio pretraining → audio–image alignment → ASR fine-tuning, with encoder weights carried forward between stages. 4.2 Approach A frozen pretrained image encoder produces the image representations, which are precomputed once and cached on disk. The audio encoder is then trained to align its own representations with these image embeddings; all 17 FastConformer blocks are updated during this stage… view at source ↗
Figure 3
Figure 3. Average WER for all 49 Indic languages evaluated, ordered by the mean across the four [PITH_FULL_IMAGE:figures/full_fig_p010_3.png] view at source ↗

Review history (2 revisions) →

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

29 extracted references · 7 linked inside Pith

  1. [7]

    Vaani: Capturing the language landscape for an inclusive digital india.arXiv preprint arXiv:2603.28714, 2026

    Sujith Pulikodan, Abhayjeet Singh, Agneedh Basu, Nihar Desai, Pranav D Bhat, Raghu Dharmaraju, Ritika Gupta, Sathvik Udupa, Saurabh Kumar, Sumit Sharma, et al. Vaani: Capturing the language landscape for an inclusive digital india.arXiv preprint arXiv:2603.28714, 2026

    arXiv 2026

  2. [1]

    Robust Speech Recognition via Large-Scale Weak Supervision

    Alec Radford, Jong Wook Kim, Tao Xu, Greg Brockman, Christine McLeavey, and Ilya Sutskever. Robust Speech Recognition via Large-Scale Weak Supervision. InProceedings of the 40th International 11 Conference on Machine Learning (ICML), volume 202 ofProceedings of Machine Learning Research, pages 28492–28518, 2023

    2023

  3. [2]

    Google USM: Scaling Automatic Speech Recognition Beyond 100 Languages

    Yu Zhang, Wei Han, James Qin, Yongqiang Wang, Ankur Bapna, Zhehuai Chen, Nanxin Chen, Bo Li, Vera Axelrod, Gary Wang, et al. Google USM: Scaling Automatic Speech Recognition Beyond 100 Languages. InarXiv preprint arXiv:2303.01037, 2023

    arXiv 2023

  4. [3]

    XLS-R: Self-supervised Cross-lingual Speech Representation Learning at Scale.arXiv preprint arXiv:2111.09296, 2022

    Arun Babu, Changhan Wang, Andros Tjandra, Kushal Lakhotia, Qiantong Xu, Naman Goyal, Kritika Singh, Patrick von Platen, Yatharth Saraf, Juan Pino, Alexei Baevski, Armand Joulin, and Michael Auli. XLS-R: Self-supervised Cross-lingual Speech Representation Learning at Scale.arXiv preprint arXiv:2111.09296, 2022

    arXiv 2022

  5. [4]

    Indicconformer-600m-multilingual

    AI4Bharat. Indicconformer-600m-multilingual. https://huggingface.co/ai4bharat/ indic-conformer-600m-multilingual , 2025. Hugging Face model repository. Accessed: 2026-08-04

    2025

  6. [5]

    Saaras: State-of-the-Art Indic ASR

    Sarvam AI. Saaras: State-of-the-Art Indic ASR. https://www.sarvam.ai/blogs/saaras, 2024. Sarvam Saaras v3, accessed 2025

    2024

  7. [6]

    Fast conformer with linearly scalable attention for efficient speech recognition

    Dima Rekesh, Nithin Rao Koluguri, Samuel Kriman, Somshubra Majumdar, Vahid Noroozi, He Huang, Oleksii Hrinchuk, Krishna Puvvada, Ankur Kumar, Jagadeesh Balam, et al. Fast conformer with linearly scalable attention for efficient speech recognition. In2023 IEEE Automatic Speech Recognition and Understanding Workshop (ASRU), pages 1–8. IEEE, 2023

    2023

  8. [8]

    wav2vec 2.0: A Framework for Self-Supervised Learning of Speech Representations

    Alexei Baevski, Henry Zhou, Abdelrahman Mohamed, and Michael Auli. wav2vec 2.0: A Framework for Self-Supervised Learning of Speech Representations. InAdvances in Neural Information Processing Systems (NeurIPS), volume 33, pages 12449–12460, 2020

    2020

Show all 29 references
  1. [9]

    Common voice: A massively-multilingual speech corpus

    Rosana Ardila, Megan Branson, Kelly Davis, Michael Kohler, Josh Meyer, Michael Henretty, Reuben Morais, Lindsay Saunders, Francis Tyers, and Gregor Weber. Common voice: A massively-multilingual speech corpus. InProceedings of the twelfth language resources and evaluation confe...

    2020

  2. [10]

    Fleurs: Few-shot learning evaluation of universal representations of speech

    Alexis Conneau, Min Ma, Simran Khanuja, Yu Zhang, Vera Axelrod, Siddharth Dalmia, Jason Riesa, Clara Rivera, and Ankur Bapna. Fleurs: Few-shot learning evaluation of universal representations of speech. In 2022 IEEE Spoken Language Technology Workshop (SLT), pages 798–805. IEEE, 2023

    2022

  3. [11]

    Indic TTS: A text-to-speech database for indian languages, 2023

    Speech Technology Consortium, Hema A Murthy, and S Umesh. Indic TTS: A text-to-speech database for indian languages, 2023. URLhttps://www.iitm.ac.in/donlab/indictts/

    2023

  4. [12]

    Indicsuperb: A speech processing universal performance benchmark for indian languages

    Tahir Javed, Kaushal Bhogale, Abhigyan Raman, Pratyush Kumar, Anoop Kunchukuttan, and Mitesh M Khapra. Indicsuperb: A speech processing universal performance benchmark for indian languages. In Proceedings of the AAAI conference on artificial intelligence, volume 37, pages 1294...

    2023

  5. [13]

    Respin-s1.0: A read speech corpus of 10000+ hours in dialects of nine indian languages

    Saurabh Kumar, Abhayjeet Singh, DEEKSHITHA G, Amartya veer, Jesuraj Bandekar, Savitha Murthy, Sumit Sharma, Sandhya Badiger, Sathvik Udupa, Amala Nagireddi, Srinivasa Raghavan K M, Rohan Saxena, Jai Nanavati, Raoul Nanavati, Janani Sridharan, Arjun Mehta, Ashish S, Sai Mora, P...

    2025

  6. [14]

    Gram vaani asr challenge on spontaneous telephone speech recordings in regional variations of hindi

    Anish Bhanushali, Grant Bridgman, Prasanta Ghosh, Pratik Kumar, Saurabh Kumar, Adithya Raj Kolladath, Nithya Ravi, Aaditeshwar Seth, Ashish Seth, Abhayjeet Singh, et al. Gram vaani asr challenge on spontaneous telephone speech recordings in regional variations of hindi. InProc...

    2022

  7. [15]

    Multilingual and code-switching asr challenges for low resource indian languages.arXiv preprint arXiv:2104.00235, 2021

    Anuj Diwan, Rakesh Vaideeswaran, Sanket Shah, Ankita Singh, Srinivasa Raghavan, Shreya Khare, Vinit Unni, Saurabh Vyas, Akash Rajpuria, Chiranjeevi Yarra, et al. Multilingual and code-switching asr challenges for low resource indian languages.arXiv preprint arXiv:2104.00235, 2021

    2021 arXiv

  8. [16]

    Gemini: A Family of Highly Capable Multimodal Models

    Google DeepMind. Gemini: A Family of Highly Capable Multimodal Models. https://deepmind. google/technologies/gemini/, 2024. Gemini 3 Flash, accessed 2025. 12

    2024

  9. [17]

    M. Harper. NeMo: a toolkit for building AI applications using Neural Modules. NVIDIA, 2019. https: //github.com/NVIDIA/NeMo

    2019

  10. [18]

    Conformer: Convolution-augmented Transformer for Speech Recognition

    Anmol Gulati, James Fan, Chung-Cheng Chiu, Niki Parmar, Yu Zhang, Jiahui Yu, Wei Han, Shibo Wang, Zhengdong Zhang, Yonghui Wu, and Ruoming Pang. Conformer: Convolution-augmented Transformer for Speech Recognition. InProceedings of Interspeech, pages 5036–5040, 2020. doi: 10.21...

    2020 doi

  11. [19]

    Park, William Chan, Yu Zhang, Chung-Cheng Chiu, Barret Zoph, Ekin D

    Daniel S. Park, William Chan, Yu Zhang, Chung-Cheng Chiu, Barret Zoph, Ekin D. Cubuk, and Quoc V . Le. SpecAugment: A Simple Data Augmentation Method for Automatic Speech Recognition. InProceedings of Interspeech, 2019

    2019

  12. [20]

    Sigmoid loss for language image pre-training

    Xiaohua Zhai, Basil Mustafa, Alexander Kolesnikov, and Lucas Beyer. Sigmoid loss for language image pre-training. In2023 IEEE/CVF International Conference on Computer Vision (ICCV), pages 11941–11952. IEEE, 2023

    2023

  13. [21]

    Colbert: Efficient and effective passage search via contextualized late interaction over bert

    Omar Khattab and Matei Zaharia. Colbert: Efficient and effective passage search via contextualized late interaction over bert. InSIGIR, 2020

    2020

  14. [22]

    Siglip 2: Multilingual vision-language encoders with improved semantic understanding, localization, and dense features, 2025

    Michael Tschannen, Alexey Gritsenko, Xiao Wang, Muhammad Ferjad Naeem, Ibrahim Alabdulmohsin, Nikhil Parthasarathy, Talfan Evans, Lucas Beyer, Ye Xia, Basil Mustafa, Olivier Hénaff, Jeremiah Harmsen, Andreas Steiner, and Xiaohua Zhai. Siglip 2: Multilingual vision-language enc...

    2025

  15. [23]

    Efficient Sequence Transduction by Jointly Predicting Tokens and Durations

    Hainan Xu, Yangyang Shi, Mahdi Fazel-Zarandi, and Yatharth Saraf. Efficient Sequence Transduction by Jointly Predicting Tokens and Durations. InProceedings of ICASSP, 2024

    2024

  16. [24]

    Indicvoices: Towards building an inclusive multilingual speech dataset for indian languages

    Tahir Javed, Janki Nawale, Eldho George, Sakshi Joshi, Kaushal Bhogale, Deovrat Mehendale, Ishvinder Sethi, Aparna Ananthanarayanan, Hafsah Faquih, Pratiti Palit, et al. Indicvoices: Towards building an inclusive multilingual speech dataset for indian languages. InFindings of ...

    2024

  17. [25]

    Spring-inx: A multilingual indian language speech corpus by spring lab, iit madras

    Arjun Gangwar, S Umesh, Rithik Sarab, Akhilesh Kumar Dubey, Govind Divakaran, Suryakanth V Gangashetty, et al. Spring-inx: A multilingual indian language speech corpus by spring lab, iit madras. arXiv preprint arXiv:2310.14654, 2023

    arXiv

  18. [26]

    SPICOR TTS_1.0 Corpus: A 97+ Hour Domain-Rich Indian English TTS Corpus

    Abhayjeet et al. SPICOR TTS_1.0 Corpus: A 97+ Hour Domain-Rich Indian English TTS Corpus. https://spiredatasets.ee.iisc.ac.in/englishttscorpus, 2025. Dataset. Accessed: 2026-08- 04

    2025

  19. [27]

    SYSPIN_S1.0 Corpus: A TTS Corpus of 900+ Hours in Nine Indian Languages

    Abhayjeet et al. SYSPIN_S1.0 Corpus: A TTS Corpus of 900+ Hours in Nine Indian Languages. https://spiredatasets.ee.iisc.ac.in/syspincorpus, 2025. Dataset. Accessed: 2026-08-04

    2025

  20. [28]

    Subword dictionary learning and segmentation techniques for automatic speech recognition in tamil and kannada, 2022

    Madhavaraj A, Bharathi Pilar, and Ramakrishnan A G. Subword dictionary learning and segmentation techniques for automatic speech recognition in tamil and kannada, 2022. URL https://arxiv.org/ abs/2207.13331

    2022 arXiv

  21. [29]

    Knowledge-driven subword grammar modeling for automatic speech recognition in tamil and kannada, 2022

    Madhavaraj A, Bharathi Pilar, and Ramakrishnan A G. Knowledge-driven subword grammar modeling for automatic speech recognition in tamil and kannada, 2022. URL https://arxiv.org/abs/2207.13333. 13

    2022 arXiv

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.