---
library_name: transformers
license: apache-2.0
tags:
- pytorch
- whisper
- automatic-speech-recognition
base_model:
- biodatlab/whisper-th-medium-combined
language:
- th
---

# Typhoon-isan-asr-whisper

<style>
img {
 display: inline;
}
</style>

| [![Model architecture](https://img.shields.io/badge/Model_Arch-Whisper_(Encoder--Decoder)-lightgrey#model-badge)](#model-architecture)
| [![Model size](https://img.shields.io/badge/Params-769M-lightgrey#model-badge)](#model-architecture)
| [![Language](https://img.shields.io/badge/Language-th-lightgrey#model-badge)](#datasets)

**Typhoon Isan ASR Whisper** is a specialized, fine-tuned version of the [biodatlab/whisper-th-medium-combined](https://huggingface.co/biodatlab/whisper-th-medium-combined) model, optimized specifically for the Isan dialect of the Thai language. Built for high-accuracy offline transcription, it delivers state-of-the-art performance on Isan speech. This enables users to host their own ASR service for Isan dialect recognition, reducing costs and avoiding the need to send sensitive data to third-party cloud services.

The model is based on [OpenAI's Whisper architecture](https://huggingface.co/openai/whisper-medium), utilizing the robust Thai-enhanced foundation from biodatlab to ensure superior understanding of regional tones and vocabulary.

By using this model, you agree to the OpenTyphoon Terms and Conditions and acknowledge the Privacy Notice: https://opentyphoon.ai/tac · https://opentyphoon.ai/privacy

**Try our demo available on [Demo]()**

**Code / Examples available on [Github]()**

**Release Blog available on [OpenTyphoon Blog]()**

***

### Performance

<img src="https://storage.googleapis.com/typhoon-public/assets/typhoon_isan_asr/performance.png" alt="cer comparison" width="90%"/>

**Note on Baseline:** The `scb10x/whisper-medium-slscu-nectec` included in the comparison is a model we fine-tuned specifically for this benchmark using existing dialect data from NECTEC and SLSCU. It serves as a representative baseline for performance based on public data, distinct from the [SLSCU_korat_model](https://huggingface.co/SLSCU/thai-dialect_korat_model) (the prominent previous work for Isan dialect ASR). This helps to determine the clear gap between capabilities derived from previously available resources and the new Typhoon Isan ASR.

### Key Findings

*   **Outperforming Proprietary State-of-the-Art:** The `typhoon-isan-asr-whisper` model achieves a Character Error Rate (CER) of **0.0885**, surpassing **Gemini-2.5-pro** (0.1020) by a clear margin. This result validates that a specialized, fine-tuned open model can deliver superior accuracy compared to massive, general-purpose proprietary systems for the Isan dialect.
*   **The Definitive Choice for Accuracy:** Among all architectures tested—including latency-optimized realtime models and historical baselines—the Whisper-based model stands out as the absolute leader. While our `typhoon-isan-asr-realtime` model is highly competitive (0.1065), the `typhoon-isan-asr-whisper` offers the highest possible fidelity, making it the preferred choice for offline transcription where precision is paramount.
*   
## **Follow us**

**https://twitter.com/opentyphoon**

## **Support**

**https://discord.gg/us5gAYmrxw**