app.py

import os 
import sys 
os.system("pip install transformers==4.27.0")
from transformers import pipeline
import gradio as gr
os.system("pip install evaluate")
os.system("pip install datasets")
os.system("pip install llvmlite")
os.system("pip install spicy")
os.system("pip install soundfile")
os.system("pip install datasets[audio]")
#os.system("pip install numpy==1.21.4")
#os.system("pip install numpy==1.22.1")
os.system("pip install numba==0.51.2")
from evaluate import evaluator
from datasets import load_dataset, Audio
from datasets import disable_caching
from datasets import set_caching_enabled
set_caching_enabled(False)
disable_caching()

p = pipeline("automatic-speech-recognition")

task_evaluator = evaluator("automatic-speech-recognition")
#url = {"test" : "https://huggingface.co/datasets/mskov/miso_test/blob/main/test_set.parquet"}
#data = load_dataset("audiofolder", data_dir="mskov/miso_test")
# data = load_dataset("audiofolder", data_files=["datasets/mskov/miso_test/test_set/and.wav","mskov/miso_test/test_set/chew1.wav","mskov/miso_test/test_set/chew3.wav", "mskov/miso_test/test_set/chew3.wav","mskov/miso_test/test_set/chew4.wav","mskov/miso_test/test_set/cough1.wav","mskov/miso_test/test_set/cough2.wav","mskov/miso_test/test_set/cough3.wav","mskov/miso_test/test_set/hi.wav","mskov/miso_test/test_set/knock_knock.wav","mskov/miso_test/test_set/mouth_sounds1.wav","mskov/miso_test/test_set/mouth_sounds2.wav","mskov/miso_test/test_set/no.wav","mskov/miso_test/test_set/not_bad.wav","mskov/miso_test/test_set/oh_i_wish.wav","mskov/miso_test/test_set/pop1.wav","mskov/miso_test/test_set/really.wav","mskov/miso_test/test_set/sigh1.wav","mskov/miso_test/test_set/sigh2.wav","mskov/miso_test/test_set/slurp1.wav","mskov/miso_test/test_set/slurp2.wav","mskov/miso_test/test_set/sneeze1.wav","mskov/miso_test/test_set/sneeze2.wav","mskov/miso_test/test_set/so_i_did_it_again.wav"])
dataset = load_dataset("mskov/miso_test", split="test").cast_column("audio", Audio())
results = task_evaluator.compute(
    model_or_pipeline="mskov/whisper_esc50",
    #model_or_pipeline="mskov/whisper-small.en",
    data=dataset,
    input_column="audio",
    label_column="audio",
    device=None,
    strategy="simple",
    metric="wer",
)
print(results)


def transcribe(audio, state=""):
    text = p(audio)["text"]
    state += text + " "
    return state, state

gr.Interface(
    fn=transcribe, 
    inputs=[
        gr.Audio(source="microphone", type="filepath", streaming=True), 
        "state"
    ],
    outputs=[
        "textbox",
        "state"
    ],
    live=True).launch()