High quality TTS data for Javanese (OpenSLR 41)

OpenSLR 41 (High quality TTS data for Javanese) — as used in Lost in Transcription (Indonesian–Javanese track)

This archive re-hosts, unchanged, the OpenSLR 41 corpus that was used to train the Indonesian–Javanese submission of participant inyongkhafid in the Lost in Transcription challenge (Mozilla Data Collective / DrivenData, 2026). The competition requires external training data that is not already on MDC to be published on MDC; this upload fulfils that rule.

Original resource: High quality TTS data for Javanese (SLR41), https://www.openslr.org/41/ — "Multi-speaker TTS data for Javanese (jv-ID)", developed by Google in partnership with Universitas Gadjah Mada. License: Attribution-ShareAlike 4.0 (CC BY-SA 4.0).

Contents

How the data was used

• All utterances: fine-tuning of the ASR ensemble members (Qwen3-Omni-30B-A3B, Qwen3-ASR-1.7B, omniASR-LLM-7B, Whisper large-v3 LoRA), and text-only adaptation of the Qwen3-ASR members.
• 4,000 utterances (with 2,000 utterances of the MDC dataset TTS Central Javanese) were also used to fine-tune a LoRA for the VoxCPM2 TTS model, which voiced the synthetic dataset published separately by the same uploader (Synthetic Javanese–Indonesian conversational speech).
• Normalisation applied for training (transcript_as_used): spelled letters x_letter → capital letters (merged, e.g. p_letter t_letter → PT); part-of-speech disambiguation tags (_nou, _vrb, …) removed; Unicode NFC and diacritics folded; dh → d, th → t. The audio was not modified.

Licence and attribution

Released under CC BY-SA 4.0, the licence of the original corpus. Changes made: one added metadata file (metadata_as_used.tsv). No other terms are added: CC BY-SA 4.0 does not permit additional restrictions.

Please cite the original corpus:

@inproceedings{kjartansson-etal-tts-sltu2018,
  title = {{A Step-by-Step Process for Building TTS Voices Using Open Source Data and Framework for Bangla, Javanese, Khmer, Nepali, Sinhala, and Sundanese}},
  author = {Keshan Sodimana and Knot Pipatsrisawat and Linne Ha and Martin Jansche and Oddur Kjartansson and Pasindu De Silva and Supheakmungkol Sarin},
  booktitle = {Proc. The 6th Intl. Workshop on Spoken Language Technologies for Under-Resourced Languages (SLTU)},
  year = {2018}, address = {Gurugram, India}, month = {August}, pages = {66--70}
}

Samples (original transcripts)

• jvf_04982_00083400049 — kara iki isine cuma wong lima thok wedok kabeh
• jvf_07638_00299821958 — pabrikipun nabisco sakmenika mboten tebih saking griyane bulik pri
• jvm_00027_00550619562 — montor toyota iku gawe gaweane entheng lan énak
• jvm_01519_01310118341 — sapa c_letter-en e_letter-en o_letter ne repsol pas tahun rong ewu
• jvm_04285_00955421578 — sondre lerche gawe rèng rèngan sakdurunge ujian sosiologi dimulai
• jvm_08178_01897618232 — phi collins ngombe obat sakwise mangan téla karo pohung

Legal basis and notice to the original authors

Original source: https://www.openslr.org/41/ (the corpus is also distributed there). Non-exclusive.

Legal basis: OpenSLR 41 was collected by Google in collaboration with Universitas Gadjah Mada and released by Google under the Creative Commons Attribution-ShareAlike 4.0 licence (Copyright 2016, 2017, 2018 Google LLC; LICENSE file at https://www.openslr.org/41/). Section 2(a)(1) of this licence lets anyone reproduce and share the material, in whole or in part, including for commercial purposes, on the conditions of attribution (Section 3(a)) and ShareAlike (Section 3(b)). This upload meets both: it keeps the same licence, credits the original publisher (copyright notice, citation and link), and contains all 5,822 utterances unchanged. Other platforms already redistribute this corpus under the same licence; for example, it is included in the "openslr" dataset on Hugging Face, which was not published by us. On 6 October 2026, at the request of the MDC reviewers, we notified the original authors (Google) that this copy is published on MDC under the same licence.