Bolbosh: Script-Aware Flow Matching for Kashmiri Text-to-Speech
arXiv:2603.07513v1 Announce Type: new Abstract: Kashmiri is spoken by around 7 million people but remains critically underserved in speech technology, despite its official status and rich linguistic heritage. The lack of robust Text-to-Speech (TTS) systems limits digital accessibility and inclusive human-computer interaction for native speakers. In this work, we present the first dedicated open-source neural TTS system designed for Kashmiri. We show that zero-shot multilingual baselines trained for Indic languages fail to produce intelligible speech, achieving a Mean Opinion Score (MOS) of only 1.86, largely due to inadequate modeling of Perso-Arabic diacritics and language-specific phonotactics. To address these limitations, we propose Bolbosh, a supervised cross-lingual adaptation strategy based on Optimal Transport Conditional Flow Matching (OT-CFM) within the Matcha-TTS framework. This enables stable alignment under limited paired data. We further introduce a three-stage acoustic
arXiv:2603.07513v1 Announce Type: new Abstract: Kashmiri is spoken by around 7 million people but remains critically underserved in speech technology, despite its official status and rich linguistic heritage. The lack of robust Text-to-Speech (TTS) systems limits digital accessibility and inclusive human-computer interaction for native speakers. In this work, we present the first dedicated open-source neural TTS system designed for Kashmiri. We show that zero-shot multilingual baselines trained for Indic languages fail to produce intelligible speech, achieving a Mean Opinion Score (MOS) of only 1.86, largely due to inadequate modeling of Perso-Arabic diacritics and language-specific phonotactics. To address these limitations, we propose Bolbosh, a supervised cross-lingual adaptation strategy based on Optimal Transport Conditional Flow Matching (OT-CFM) within the Matcha-TTS framework. This enables stable alignment under limited paired data. We further introduce a three-stage acoustic enhancement pipeline consisting of dereverberation, silence trimming, and loudness normalization to unify heterogeneous speech sources and stabilize alignment learning. The model vocabulary is expanded to explicitly encode Kashmiri graphemes, preserving fine-grained vowel distinctions. Our system achieves a MOS of 3.63 and a Mel-Cepstral Distortion (MCD) of 3.73, substantially outperforming multilingual baselines and establishing a new benchmark for Kashmiri speech synthesis. Our results demonstrate that script-aware and supervised flow-based adaptation are critical for low-resource TTS in diacritic-sensitive languages. Code and data are available at: https://github.com/gaash-lab/Bolbosh.
Executive Summary
The article introduces Bolbosh, the first open-source neural Text-to-Speech system tailored for Kashmiri, a linguistically rich but digitally underserved language with 7 million speakers. Addressing a critical gap, the authors demonstrate that generic multilingual baselines fail to produce intelligible speech for Kashmiri due to inadequate modeling of Perso-Arabic diacritics and phonotactics, yielding a MOS of 1.86. Bolbosh leverages a supervised cross-lingual adaptation via Optimal Transport Conditional Flow Matching (OT-CFM) within Matcha-TTS, enabling stable alignment with limited paired data. Additionally, a three-stage acoustic enhancement pipeline—dereverberation, silence trimming, and loudness normalization—unifies heterogeneous sources. The system’s vocabulary expansion to encode Kashmiri graphemes preserves vowel distinctions, achieving a MOS of 3.63 and MCD of 3.73, outperforming multilingual baselines and establishing a new benchmark. The work underscores the necessity of script-aware, supervised adaptation for low-resource diacritic-sensitive languages.
Key Points
- ▸ First open-source Kashmiri TTS system developed
- ▸ Multilingual baselines fail due to diacritic and phonotactic modeling gaps
- ▸ Bolbosh introduces OT-CFM and acoustic enhancement pipeline to achieve superior performance
Merits
Script-Aware Adaptation
Bolbosh’s use of OT-CFM and explicit encoding of Kashmiri graphemes directly addresses linguistic specificity, yielding a meaningful MOS improvement from 1.86 to 3.63.
Acoustic Enhancement Pipeline
The three-stage pipeline (dereverberation, silence trimming, loudness normalization) effectively stabilizes alignment learning across heterogeneous sources, enhancing overall synthesis quality.
Demerits
Limited Paired Data Constraint
The reliance on cross-lingual adaptation with limited paired data may restrict scalability or generalizability beyond the Kashmiri-specific dataset used.
Expert Commentary
Bolbosh represents a pivotal advancement in low-resource speech synthesis by demonstrating that linguistic specificity—particularly diacritic sensitivity—cannot be ignored without compromising intelligibility. The choice to adopt OT-CFM within Matcha-TTS is particularly astute: it leverages a mathematically rigorous mechanism for cross-lingual adaptation under data scarcity, avoiding the pitfalls of generic multilingual models that ignore orthographic and phonological nuances. The acoustic enhancement pipeline, though standard in many systems, is effectively contextualized here as a critical enabler for aligning heterogeneous speech inputs, a common obstacle in real-world deployment. Importantly, the authors’ decision to expand the vocabulary to explicitly encode Kashmiri graphemes is not merely technical—it is pedagogical and representational; it signals a commitment to linguistic integrity and user dignity. This work should serve as a model for future efforts in endangered or underrepresented languages, particularly those with complex script systems. The open-source release further amplifies its impact, transforming a research contribution into a catalyst for broader accessibility.
Recommendations
- ✓ 1. Researchers working on similar low-resource languages should adopt Bolbosh’s OT-CFM framework as a baseline for cross-lingual adaptation, especially when diacritics or orthographic complexity are present.
- ✓ 2. Policymakers and NGOs supporting digital inclusion should prioritize funding and partnerships with open-source TTS initiatives like Bolbosh to ensure equitable access to speech technology for marginalized linguistic communities.