README.md

---
base_model: openai/whisper-small
library_name: peft
language:
- ja
pipeline_tag: automatic-speech-recognition
tags:
- lora
- whisper
- speech-to-text
- kansai-dialect
- japanese
---

# Whisper-Small-Kansai-LoRA

[JP] このモデルは、OpenAIの **Whisper Small** をベースに、**関西弁（近畿方言）**に特化してファインチューニングした音声認識（STT）用LoRAモデルです。  
[EN] This is a LoRA adapter for **Whisper Small**, fine-tuned specifically for the **Kansai dialect** (Kinki dialect) of Japanese.

## Model Details / モデルの詳細

### [JP] 特徴
標準的なWhisperでは標準語に修正されがちな「〜やねん」「〜しはる」といった関西弁特有の表現を、聞こえたままに文字起こしすることを目指しています。

### [EN] Features
While standard Whisper models often "correct" Kansai dialect into standard Japanese, this model aims to transcribe specific Kansai expressions (e.g., "~yanen", "~shiharu") as they are spoken.

- **Developer:** hpscript
- **Language:** Japanese (Kansai Dialect)
- **Base Model:** [openai/whisper-small](https://huggingface.co/openai/whisper-small)
- **Method:** PEFT (LoRA)

## Usage / 使い方

```python
import torch
from transformers import WhisperForConditionalGeneration, WhisperProcessor
from peft import PeftModel

base_model_id = "openai/whisper-small"
lora_model_id = "hpscript/whisper-small-kansai-lora"

model = WhisperForConditionalGeneration.from_pretrained(base_model_id)
model = PeftModel.from_pretrained(model, lora_model_id)
processor = WhisperProcessor.from_pretrained(base_model_id)