Transcript language is not specified correctly.

[Cyp9715]

First of all, the code is as follows and follows the basic preset provided by huggingface.

import torch
from transformers import AutoModelForSpeechSeq2Seq, AutoProcessor, pipeline
from datasets import load_dataset

device = "cuda:0" if torch.cuda.is_available() else "cpu"
torch_dtype = torch.float16 if torch.cuda.is_available() else torch.float32

model_id = "openai/whisper-large-v3"

model = AutoModelForSpeechSeq2Seq.from_pretrained(model_id, torch_dtype=torch_dtype, low_cpu_mem_usage=True, use_safetensors=True)
model.to(device)

processor = AutoProcessor.from_pretrained(model_id)

pipe = pipeline(
    "automatic-speech-recognition",
    model=model,
    tokenizer=processor.tokenizer,
    feature_extractor=processor.feature_extractor,
    max_new_tokens=128,
    chunk_length_s=30,
    batch_size=16,
    return_timestamps=True,
    torch_dtype=torch_dtype,
    device=device,
)

# test
result = pipe('Data.wav', generate_kwargs={"task":"transcribe", "language":"<|ko|>"})
print(result['text'])

wav file provided by AI Hub in South Korea.
There are hundreds of thousands of voice files, most of them are transcribe correctly.

but
{"task":"transcribe", "language":"<|ko|>"}

Even if specify , there is an error in recognizing some specific files (not good quality, incorrect pronunciation of the speaker) as Japanese.
Whisper outputs the following results.

• グッデナシがちんちゃんの思うとおか지고。

After testing the <|en|> option, it is successfully transcribed into English.
Whisper outputs the following results.

• I was really scared.

In conclusion, the 'ko' option does not appear to work properly.

Does anyone know the solution?