app.py

from transformers import pipeline
import gradio as gr
import torch

device = 0 if torch.cuda.is_available() else "cpu"

MODEL_NAME = "6x16/whisper-small-nan-tw-quicktrain"

pipe = pipeline(model=MODEL_NAME,
                task="automatic-speech-recognition",
                chunk_length_s=30,
                device=device)  # change to "your-username/the-name-you-picked"

pipe.model.config.forced_decoder_ids = None
# pipe.model.config.suppress_tokens = []
pipe.model.generation_config.forced_decoder_ids = None
# pipe.model.generation_config._from_model_config = True

def transcribe(inputs, task):
    if inputs is None:
        raise gr.Error("No audio file found or wait until audio input is ready.")
    text = pipe(inputs, generate_kwargs={"task": task})
    return  text["text"]

iface = gr.Interface(
    fn=transcribe,
    inputs=[gr.Audio(sources="microphone", type="filepath"),
            gr.Radio(["transcribe", "translate"], label="Task")],
    outputs="text",
    theme="glass",
    title="Whisper Small nan-tw  model by self-training (閩南話/台語)",
    description="Realtime demo for Minnan speech recognition using a self-tuned Whisper small model."
                f"\tCheckpoint: [{MODEL_NAME}](https://huggingface.co/{MODEL_NAME})",
    allow_flagging="never",
)

iface.launch()