app.py

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

#lmID = "aware-ai/german-lowercase-wiki-5gram"
#decoder = BeamSearchDecoderCTC.load_from_hf_hub(lmID)
p = pipeline("automatic-speech-recognition", model="aware-ai/wav2vec2-xls-r-1b-5gram-german")
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"]
    return transcribed
    
def punctuate(transcribed):
    punctuated = ttp(transcribed, max_length = 512)[0]["generated_text"]
    return punctuated

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

interfaces = [
    get_asr_interface(),
    get_punctuation_interface(),
]

names = [
    "ASR",
    "GRAMMAR",
]

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