Language support

[dgoryeo]

Does FunASR support transcription of Japanese language?
Thanks

[LauraGPT]

Yes. Current FunASR has two practical Japanese transcription paths:

1. SenseVoiceSmall (CPU or GPU)

SenseVoiceSmall is the lighter 234M model and supports Japanese explicitly. Set language="ja" when the input is known to be Japanese, or use "auto" for automatic language detection.

from funasr import AutoModel
from funasr.utils.postprocess_utils import rich_transcription_postprocess

model = AutoModel(
    model="iic/SenseVoiceSmall",
    vad_model="fsmn-vad",
    device="cpu",  # or "cuda"
)

result = model.generate(
    input="japanese.wav",
    language="ja",
    use_itn=True,
    batch_size_s=60,
)
print(rich_transcription_postprocess(result[0]["text"]))

2. Fun-ASR-Nano (GPU)

Fun-ASR-Nano supports Chinese, English, and Japanese and is the current flagship LLM-ASR path:

from funasr import AutoModel

model = AutoModel(
    model="FunAudioLLM/Fun-ASR-Nano-2512",
    device="cuda",
)

result = model.generate(input="japanese.wav")
print(result[0]["text"])

Do not use paraformer-zh for Japanese; its listed language coverage is Chinese and English only.

Current source references:

• Model language table
• FunASR model-selection guide
• SenseVoice language options
• SenseVoice ja language mapping