Home / Speech / listen-windy-lingua-ps

Pashto speech recognition model listen-windy-lingua-ps

Open speech-to-text model published by Windstorm Labs on Hugging Face.

Task: Speech-to-text (automatic speech recognition)
Language: Pashto
Lineage: Whisper community fine-tune
Formats in the repo: safetensors
Hugging Face: WindyWord/listen-windy-lingua-ps
Last updated on HF: 2026-05-06

Quality

Not benchmarked by this site yet. Speech models have no published word-error rates here; any figures in the Hugging Face card were not re-measured by us.

From the model card: EXCELLENT tier when used correctly. Derived from `ihanif/whisper-medium-pashto`. Verified at WER 5.3% / CER 3.2% / script-match 99.2% on 50-sample FLEURS ps_af *when inference uses* `forced_decoder_ids` (passed explicitly to `model.generate()` via `processor.get_decoder_prompt_ids(language='pashto', task='transcribe')`). With the convenience `language=` kwarg the model can silently drop the Pashto token and hallucinate English script on ~30% of samples (53.7% WER artifact). Always force the decoder prompt for Pashto inference.

How to run

Shell

pip install transformers torch huggingface_hub   # plus ffmpeg on your PATH to read audio files

Python

from huggingface_hub import snapshot_download
from transformers import pipeline

# Weights live in the safetensors/ subfolder of the repository.
path = snapshot_download("WindyWord/listen-windy-lingua-ps", allow_patterns=["safetensors/*"])
asr = pipeline("automatic-speech-recognition", model=f"{path}/safetensors")
print(asr("audio.wav")["text"])

Tested: Pattern run on 2026-10-01 against WindyWord/listen-windy-nano with transformers 4.57.6 and 5.18.0 on CPU, on a 16 kHz mono WAV. Not run on every model individually.

Licence and attribution

Licence: Apache-2.0 lineage under review
Upstream: ihanif/whisper-medium-pashto (model card text)
Upstream licence: Apache-2.0
Attribution: Based on ihanif/whisper-medium-pashto, a community fine-tune of OpenAI Whisper. Licensed Apache-2.0. Lineage under review.

Lineage under review: this model was imported from a community Whisper fine-tune (ihanif/whisper-medium-pashto). Some community fine-tunes carry non-commercial terms, so we cannot yet confirm the Apache-2.0 tag. Check before commercial use.

Copy this into your NOTICE file (or equivalent) when you ship the model:

NOTICE

This product uses the model WindyWord/listen-windy-lingua-ps
(https://huggingface.co/WindyWord/listen-windy-lingua-ps), published by Windstorm Labs (https://windstormlabs.com).
Based on ihanif/whisper-medium-pashto, a community fine-tune of OpenAI Whisper. Licensed Apache-2.0. Lineage under review.
Imported by Windstorm Labs from an upstream community fine-tune; see the model card.
Licensed under the Apache License, Version 2.0: https://www.apache.org/licenses/LICENSE-2.0

This is a convenience, not legal advice. See Licensing.