Model reference · open weights
monsoon-whisper-medium-gigaspeech2 is an open-weight audio or speech model from typhoon-ai. monsoon-whisper-medium-gigaspeech2 (FP32) weighs 1.5 GB; the smallest configuration that runs it is RTX 3060 12 GB.
For self-hosting, compare this build's requirements below with the GPU you plan to rent.
Summary of the typhoon-ai/monsoon-whisper-medium-gigaspeech2 model card, 2026-10-01
What it is
| Released by | typhoon-ai |
|---|---|
| Released | 2024-07-12 |
| Parameters | 764M |
| VRAM | 1.5 GB for the weights |
What it runs on
| Card | One stream | Memory |
|---|---|---|
| RTX 3060 12 GB | fits | 11.6 GB |
| RTX 4060 Ti 16 GB | fits | 15.4 GB |
| RTX 3090 24 GB | fits | 23.4 GB |
| RTX 4090 24 GB | fits | 23.4 GB |
| RTX 5090 32 GB | fits | 31.0 GB |
| L40S 48 GB | fits | 44.0 GB |
| A100 80 GB | fits | 78.2 GB |
| H100 80 GB | fits | 78.1 GB |
| RTX PRO 6000 Blackwell 96 GB | fits | 93.8 GB |
| DGX Spark (GB10) 128 GB unified | fits | 107 GB |
| H200 141 GB | fits | 138 GB |
| B200 180 GB | fits | 176 GB |
From the model card
Monsoon-Whisper-Medium-GigaSpeech2 is a 🇹🇭 Thai Automatic Speech Recognition (ASR) model. It is based on Whisper-Medium and fine-tuned on GigaSpeech2.
Originally developed as a scale experiment for research on emergent capabilities in ASR tasks. It performs well in the wild, including with audio sourced from YouTube and in noisy environments.
More details can be found in our Typhoon-Audio Release Blog.
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
from transformers import WhisperProcessor, WhisperForConditionalGeneration
import torchaudio
import torch
model_path = "scb10x/monsoon-whisper-medium-gigaspeech2"
device = "cuda"
filepath = 'audio.wav'
processor = WhisperProcessor.from_pretrained(model_path)
model = WhisperForConditionalGeneration.from_pretrained(
model_path, torch_dtype=torch.bfloat16
)
model.to(device)
model.eval()
model.config.forced_decoder_ids = processor.get_decoder_prompt_ids(
language="th", task="transcribe"
)
array, sr = torchaudio.load(filepath)
input_features = (
processor(array, sampling_rate=sr, return_tensors="pt")
.to(device)
.to(torch.bfloat16)
.input_features
)
predicted_ids = model.generate(input_features)
transcription = processor.batch_decode(predicted_ids, skip_special_tokens=True)
print(transcription)
| Model | WER (GS2) | WER (CV17) | CER (GS2) | CER (CV17) |
|---|---|---|---|---|
| whisper-large-v3 | 37.02 | 22.63 | 24.03 | 8.49 |
| whisper-medium | 55.64 | 43.01 | 37.55 | 16.41 |
| biodatlab-whisper-th-medium-combined | 31.00 | 14.25 | 21.20 | 5.69 |
| biodatlab-whisper-th-large-v3-combined | 29.02 | 15.72 | 19.96 | 6.32 |
| monsoon-whisper-medium-gigaspeech2 | 22.74 | 20.79 | 14.15 | 6.92 |
This model is experimental and may not always be accurate. Developers should carefully assess potential risks in the context of their specific applications.
Kunat Pipatanakul, Potsawee Manakul, Sittipong Sripaisarnmongkol, Natapong Nitarach, Warit Sirichotedumrong, Adisai Na-Thalang, Phatrasek Jirabovonvisut, Parinthapat Pengpun, Krisanapong Jirayoot, Pathomporn Chokchainant, Kasima Tharnpipitchai
Quoted from the model card on Hugging Face. The full card is behind the Hugging Face link above.