Downloads · 30 days
0
ai4good-labyrinth/fleurs-synthetic-hours0p25-whisper-medium-no-language-lora-adapter
fleurs-synthetic-hours0p25-whisper-medium-no-language-lora-adapter is a automatic speech recognition model from ai4good-labyrinth. Use it when you need speech turned into text. The card lists the license as apache-2.0.
This repository contains a Whisper checkpoint for Chichewa/Nyanja automatic speech recognition, fine-tuned from openai/whisper-medium.
Downloads · 30 days
0
Access
Public
Updated May 27, 2026
Repo size
563 MB
Likes
0
Public
Click a slice to open those files.
.safetensors563 MB · 99%
From the Hugging Face model README
This repository contains a Whisper checkpoint for Chichewa/Nyanja
automatic speech recognition, fine-tuned from openai/whisper-medium.
fleurs-syntheticopenai/whisper-mediumfleurs_synthetic_hours0p25This adapter is intended for research and evaluation on Chichewa/Nyanja ASR. It must be used together with the base Whisper model. It is not a production-ready speech system and should be validated carefully before downstream use.
This repository contains an adapter, not a fully merged standalone model. Load the base model first, then attach the adapter.
from transformers import AutoModelForSpeechSeq2Seq, AutoProcessor
from peft import PeftModel
base_model_id = "openai/whisper-medium"
adapter_repo_id = "ai4good-labyrinth/fleurs-synthetic-hours0p25-whisper-medium-no-language-lora-adapter"
processor = AutoProcessor.from_pretrained(base_model_id)
base_model = AutoModelForSpeechSeq2Seq.from_pretrained(base_model_id)
model = PeftModel.from_pretrained(base_model, adapter_repo_id)
The local evaluation script in this repository can also load the adapter
directly because it reads adapter_config.json and automatically fetches the
base model.
FLEURS train + Prepared datasetFLEURS dev45364495311305min_duration_seconds=0.0, max_duration_seconds=30.0experiments/whisper_finetune/finetune_whisper.pyopenai/whisper-mediumtranscribe--language auto in standalone evaluationyes32640.05q_proj,out_proj,v_proj,fc1,k_proj,fc2embed_tokens,proj_outfp16True18003.20The best checkpoint was selected using trainer-side dev evaluation on the duration-filtered FLEURS dev split.
0.47920.12430.6220These values come from the training pipeline and may differ slightly from standalone post-hoc evaluation because the decoding path is not perfectly identical.
Standalone evaluation is recommended for the final release. Filtered and unfiltered results should be reported separately.
min_duration_seconds=0, max_duration_seconds=30transcribeauto| Dataset | Split | Setting | Num examples | WER | CER | Notes |
|---|---|---|---|---|---|---|
| FLEURS | dev | filtered | TBD | TBD | TBD | Standalone eval pending |
| FLEURS | dev | unfiltered | TBD | TBD | TBD | Standalone eval pending |
| FLEURS | test | filtered | TBD | TBD | TBD | Standalone eval pending |
| FLEURS | test | unfiltered | TBD | TBD | TBD | Standalone eval pending |
| Zambezi | dev | filtered | TBD | TBD | TBD | Standalone eval pending |
| Zambezi | dev | unfiltered | TBD | TBD | TBD | Standalone eval pending |
| Zambezi | test | filtered | TBD | TBD | TBD | Standalone eval pending |
| Zambezi | test | unfiltered | TBD | TBD | TBD | Standalone eval pending |
Detailed JSON metrics can be stored under:
eval/
eval/...If you use this checkpoint, please cite:
@misc{fleurs_synthetic_hours0p25_whisper_medium_no_language_lora_adapter_2026,
title = {Fleurs Synthetic Whisper Medium Fleurs Synthetic Hours0P25 LoRA Adapter},
author = {AI4Good Labyrinth Team},
year = {2026},
howpublished = {\url{https://huggingface.co/ai4good-labyrinth/fleurs-synthetic-hours0p25-whisper-medium-no-language-lora-adapter}},
note = {Whisper LoRA adapter fine-tuning for Chichewa/Nyanja ASR}
}