app.py

from transformers import pipeline
import gradio as gr
from pyctcdecode import BeamSearchDecoderCTC

lmID = "aware-ai/german-lowercase-5gram-kenlm"
decoder = BeamSearchDecoderCTC.load_from_hf_hub(lmID)
p = pipeline("automatic-speech-recognition", model="aware-ai/robust-wav2vec2-xls-r-300m-german-lowercase", decoder=decoder)
ttp = pipeline("text2text-generation", model="aware-ai/marian-german-grammar")

def transcribe(audio):
    transcribed = p(audio, chunk_length_s=16, stride_length_s=(4, 0))["text"]
    
    punctuated = ttp(transcribed, max_length = 512)[0]["generated_text"]
    
    return transcribed, punctuated 

def get_asr_interface():
    return gr.Interface(
        fn=transcribe, 
        inputs=[
            gr.inputs.Audio(source="microphone", type="filepath")
        ],
        outputs=[
            "textbox",
            "textbox"
        ])

interfaces = [
    get_asr_interface(),
]

names = [
    "ASR",
]

gr.TabbedInterface(interfaces, names).launch(server_name = "0.0.0.0", enable_queue=False)