jshrdt/lowhipa-base-asc

lowhipa-base-asc

This Whisper-for-IPA (WhIPA) model adapter is a PEFT LoRA fine-tuned version of openai/whisper-base on a subset (1k samples) of the Arabic Speech Corpus (https://en.arabicspeechcorpus.com) with custom IPA transcriptions transliterated from the provided Buckwalter transcriptions; ASC-IPA dataset available at https://doi.org/10.5281/zenodo.17111977.

Model description

For deployment and description, please refer to https://github.com/jshrdt/whipa.

from transformers import WhisperForConditionalGeneration, WhisperTokenizer, WhisperProcessor
from peft import PeftModel

tokenizer = WhisperTokenizer.from_pretrained("openai/whisper-base", task="transcribe")
tokenizer.add_special_tokens({"additional_special_tokens": ["<|ip|>"] + tokenizer.all_special_tokens})

base_model = WhisperForConditionalGeneration.from_pretrained("openai/whisper-base")
base_model.generation_config.lang_to_id["<|ip|>"] = tokenizer.convert_tokens_to_ids(["<|ip|>"])[0]
base_model.resize_token_embeddings(len(tokenizer))

whipa_model = PeftModel.from_pretrained(base_model, "jshrdt/lowhipa-base-asc")

whipa_model.generation_config.language = "<|ip|>"
whipa_model.generation_config.task = "transcribe"

whipa_processor = WhisperProcessor.from_pretrained("openai/whisper-base", task="transcribe")

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• —learning_rate: 0.001
• —trainbatchsize: 16
• —evalbatchsize: 8
• —seed: 42
• —optimizer: Use OptimizerNames.ADAMWTORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments
• —lrschedulertype: linear
• —lrschedulerwarmup_ratio: 0.1
• —num_epochs: 10
• —mixedprecisiontraining: Native AMP

Training results

Framework versions

• —PEFT 0.15.1
• —Transformers 4.48.3
• —Pytorch 2.6.0+cu124
• —Datasets 3.2.0
• —PEFT 0.15.1