Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Dialect Matters: Cross-Lingual ASR Transfer for Low-Resource Indic Language Varieties

Akriti Dhasmana, Aarohi Srivastava, David Chiang

Abstract

We conduct an empirical study of cross-lingual transfer using spontaneous, noisy, and code-mixed speech across a wide range of Indic dialects and language varieties. Our results indicate that although ASR performance is generally improved with reduced phylogenetic distance between languages, this factor alone does not fully explain performance in dialectal settings. Often, fine-tuning on smaller amounts of dialectal data yields performance comparable to fine-tuning on larger amounts of phylogenetically-related, high-resource standardized languages. We also present a case study on Garhwali, a low-resource Pahari language variety, and evaluate multiple contemporary ASR models. Finally, we analyze transcription errors to examine bias toward pre-training languages, providing additional insight into challenges faced by ASR systems on dialectal and non-standardized speech.

Dialect Matters: Cross-Lingual ASR Transfer for Low-Resource Indic Language Varieties

Abstract

We conduct an empirical study of cross-lingual transfer using spontaneous, noisy, and code-mixed speech across a wide range of Indic dialects and language varieties. Our results indicate that although ASR performance is generally improved with reduced phylogenetic distance between languages, this factor alone does not fully explain performance in dialectal settings. Often, fine-tuning on smaller amounts of dialectal data yields performance comparable to fine-tuning on larger amounts of phylogenetically-related, high-resource standardized languages. We also present a case study on Garhwali, a low-resource Pahari language variety, and evaluate multiple contemporary ASR models. Finally, we analyze transcription errors to examine bias toward pre-training languages, providing additional insight into challenges faced by ASR systems on dialectal and non-standardized speech.
Paper Structure (28 sections, 3 figures, 10 tables)

This paper contains 28 sections, 3 figures, 10 tables.

Table of Contents

  1. Introduction
  2. Contributions
  3. Related Work
  4. Languages and Data
  5. Experiments and Results
  6. Metrics
  7. Analyzing Broader Patterns in Indic ASR
  8. RQ1: Baseline Assessment
  9. Results
  10. RQ2: Orthographic Consistency
  11. Results
  12. RQ3: Cross-Lingual Transfer
  13. Results
  14. Cross-Lingual Transfer Trends
  15. Dialectal Transfer Effects
  16. ...and 13 more sections

Figures (3)

  • Figure 1: Subset of the Indo-European language family tree showing Devanagari-script Indic languages in the VAANI dataset, based on Glottolog indo1320. Languages are annotated with WER for IndicWav2Vec-Hindi. Blue highlights indicate languages used during pre-training.
  • Figure 2: Hapax legomena (once-seen word types) % per language in the test split of VAANI plotted against WER (Pearson's $\rho = 0.705$, $p = 4 \times 10^{-4}$) and CER (not significant) using IndicWav2Vec-Hindi.
  • Figure 3: Cross-Lingual Performance of w2vBERT models fine-tuned on 1 to 7 hours of data per language vs. off-the-shelf IndicWav2Vec-Hindi.