Pashto Common Voice: Building the First Open Speech Corpus for a 60-Million-Speaker Low-Resource Language

Hanif Rahman; Shafeeq ur Rehman

cs.CLMar 27, 2026

Pashto Common Voice: Building the First Open Speech Corpus for a 60-Million-Speaker Low-Resource Language

Authors: Hanif Rahman, Shafeeq ur Rehman

Organizations: Pashto DAO

Abstract

We present the Pashto Common Voice corpus -- the first large-scale, openly licensed speech resource for Pashto, a language with over 60 million native speakers largely absent from open speech technology. Through a community effort spanning 2022-2025, the corpus grew from 1.5 hours and 5 contributors to 147 total hours and 1,483 unique speakers across ten Mozilla Common Voice releases (CV14-CV23). Speaker participation increased approximately 108-fold between CV17 and CV18, coinciding with a VOA Pashto broadcast campaign. We describe the full methodology: interface localisation, Wikipedia-based sentence extraction with automated filtering, phonemically targeted contributions for the four most frequently dropped Pashto characters, and multi-channel community outreach. MCV23 contains 107,781 clips (60,337 validated; 82.33 validated hours) across 13 content domains. Fine-tuning Whisper Base on the MCV20 yields 13.4% WER on the MCV20 test split, against the published Whisper Base zero-shot WER of 99.0% on Pashto.

Figures & tables

Figure 1: Pashto Common Voice corpus growth across MCV releases 14–23, showing total hours (bars) and unique speakers (line). The CV17

\!\to

CV18 inflection (Mar.–Jun. 2024) is attributable to the VOA Pashto media campaign.

Release	Date	Hours	Speakers
CV14	Jun. 2023	1.49	5
CV15	Sep. 2023	1.66	7
CV16	Dec. 2023	1.71	8
CV17	Mar. 2024	2.11	9
CV18	Jun. 2024	75.94	971
CV19	Sep. 2024	89.72	1,098

Table 1: Pashto MCV corpus release history.

Statistic	Value
Total clips	107,781
Validated clips	60,337
Validated hours	82.33 h
Total hours	147.07 h
Unique speakers	1,483
Validated sentences	25,109

Table 2: MCV23 Pashto corpus statistics.

Figure 2: MCV23 Pashto: validated clips by content domain.

Figure 3: MCV23 speaker age distribution. Gender is unreported for 98% of speakers (a critical metadata gap).

Step	Epoch	WER	WER ortho
100	0.1	93.5	94.6
500	0.3	57.0	60.4
1000	0.6	46.7	50.6
2000	1.3	30.9	34.5
3000	1.9	22.5	25.3
4000	2.6	16.5	18.3

Table 3: WER (%) of ps_base_l1 on the MCV20 Pashto test split. Step 100 is the first logged checkpoint (not zero-shot). The Whisper Base zero-shot WER on Pashto is 99.0% on Fleurs [ 9 ] ; our fine-tuned model is evaluated on the MCV20 test split.

Speech Corpus Construction

See all 24 papers

Fine-tuning Whisper for Pashto ASR: strategies and scale
Apr 7, 2026Hanif RahmanAutomatic Speech Recognition Low-Resource Speech Recognition
Independent Researcher.
ParsVoice: A Large-Scale Multi-Speaker Persian Speech Corpus for Text-to-Speech Synthesis
Oct 12, 2025Mohammad Javad Ranjbar Kalahroodi, Heshaam Faili, Azadeh ShakeryTTS Synthesis Multi-Speaker TTS
School of Electrical and Computer Engineering, University of Tehran, Iran · Institute for Research in Fundamental Sciences (IPM), Tehran, Iran
PashtoCorp: A 1.25-Billion-Word Corpus, Evaluation Suite, and Reproducible Pipeline for Low-Resource Language Development
Mar 17, 2026Hanif RahmanLow-Resource Language Processing Language Model Pretraining
Independent Researcher

Apr 7, 2026J/K move · Enter open · S save

Fine-tuning Whisper for Pashto ASR: strategies and scale

Hanif Rahman

Independent Researcher.

Automatic Speech Recognition Low-Resource Speech Recognition Parameter-Efficient Fine-Tuning

Open paper AskPDF arXiv

Pashto is absent from Whisper's pre-training corpus despite being one of CommonVoice's largest language collections, leaving off-the-shelf models unusable: all Whisper sizes output Arabic, Dari, or Urdu script on Pashto audio, achieving word error rates above 100%. We compare four fine-tuning strategies for whisper-base on CommonVoice Pashto v20: vanilla full fine-tuning, LoRA (rank 64), frozen-encoder (2/6 layers), and multistage Urdu-to-Pashto transfer. We extend vanilla fine-tuning to whisper-small and whisper-large-v3-turbo on CommonVoice Pashto v24 (113 hours). Vanilla fine-tuning achieves WER 21.22% on CV20, outperforming LoRA by 33.36 pp, frozen-encoder by 14.76 pp, and Urdu transfer by 44.56 pp. Frozen-encoder fine-tuning degrades performance on whisper-base (6 encoder layers): layer-function separation does not hold at this depth, and freezing removes a third of trainable capacity. Urdu-to-Pashto transfer fails due to an unverified intermediate checkpoint, phonological mismatch, and insufficient training. On CV24, whisper-small achieves WER 24.89% (2.24 pp over whisper-base at 3.3x parameters); whisper-large-v3-turbo achieves 23.37% (a further 1.52 pp). Diminishing returns indicate whisper-small is the practical optimum at 113 hours. Online augmentation provides 7.25 pp WER benefit over matched training. Error analysis identifies word-final suffix confusion (masculine -ay vs. feminine -a) and retroflex substitutions involving the Pashto-unique consonant /ts/ as dominant failure modes. Fine-tuned checkpoints and evaluation scripts are released on HuggingFace.

Abstract

Speech Corpus Construction

Explore similar work

Fine-tuning Whisper for Pashto ASR: strategies and scale

ParsVoice: A Large-Scale Multi-Speaker Persian Speech Corpus for Text-to-Speech Synthesis

PashtoCorp: A 1.25-Billion-Word Corpus, Evaluation Suite, and Reproducible Pipeline for Low-Resource Language Development