app.py

import os
if os.environ.get("SPACES_ZERO_GPU") is not None:
    import spaces
else:
    class spaces:
        @staticmethod
        def GPU(func):
            def wrapper(*args, **kwargs):
                return func(*args, **kwargs)
            return wrapper
import gradio as gr
import subprocess

#subprocess.run("git clone https://github.com/AI4Bharat/NeMo.git && cd NeMo && git checkout nemo-v2 && bash reinstall.sh", shell=True)

import torch
import nemo.collections.asr as nemo_asr

from pathlib import Path

model = nemo_asr.models.ASRModel.from_pretrained("ai4bharat/indicconformer_stt_ml_hybrid_rnnt_large")

device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model.freeze() # inference mode
model = model.to(device) # transfer model to device

@spaces.GPU
def infer(srcfile: str):
    tmpfile = "sample_audio_infer_ready.wav"

    subprocess.run(f"ffmpeg -i {srcfile} -ac 1 -ar 16000 {tmpfile}", shell=True)
    model.cur_decoder = "ctc"
    ctc_text = model.transcribe([tmpfile], batch_size=1, logprobs=False, language_id='ml')[0]
    print(ctc_text)

    model.cur_decoder = "rnnt"
    rnnt_text = model.transcribe([tmpfile], batch_size=1, language_id='ml')[0]
    print(rnnt_text)

    if Path(tmpfile).exists(): Path(tmpfile).unlink()

    return ctc_text, rnnt_text

with gr.Blocks() as demo:
    input_audio = gr.Audio(label="Input", type="filepath", sources=["upload", "microphone"], format="wav")
    run_button = gr.Button("Run", variant="primary")
    with gr.Row():
        ctc_text = gr.Textbox(label="CTC", value="", show_copy_button=True)
        rnnt_text = gr.Textbox(label="RNNT", value="", show_copy_button=True)

    run_button.click(infer, [input_audio], [ctc_text, rnnt_text])

demo.launch()