app.py

import os 
import sys 
os.system("pip install transformers==4.27.0")
os.system("pip install torch")
os.system("pip install openai")
os.system("pip install accelerate")
from transformers import pipeline, WhisperModel, WhisperTokenizer, WhisperFeatureExtractor
os.system("pip install evaluate")
#import evaluate
#os.system("pip install evaluate[evaluator]")
os.system("pip install datasets")
# os.system("pip install llvmlite")
os.system("pip install spicy==1.8.1")
os.system("pip install soundfile")
os.system("pip install jiwer")
os.system("pip install datasets[audio]")
os.system("pip install numba==0.51.2")
from evaluate import evaluator
from datasets import load_dataset, Audio, disable_caching, set_caching_enabled

set_caching_enabled(False)
disable_caching()


#config = AutoConfig.from_pretrained('whisper-small')

huggingface_token = os.environ["huggingface_token"]

miso_tokenizer = WhisperTokenizer.from_pretrained("mskov/whisper_miso", use_auth_token=huggingface_token)
miso_feature_extractor = WhisperFeatureExtractor.from_pretrained("mskov/whisper_miso", use_auth_token=huggingface_token)
whisper_miso=WhisperModel.from_pretrained("mskov/whisper_miso", use_auth_token=huggingface_token, tokenizer=miso_tokenizer, device_map="auto")

#whisper_miso=WhisperModel.from_pretrained("openai/whisper-small", use_auth_token=huggingface_token, device_map="auto")




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())
# dataset = load_dataset("mskov/miso_test", split="test")
dataset = load_dataset("mskov/ESC50", split="test")
print(dataset)
results = task_evaluator.compute(
    model_or_pipeline=whisper_miso,
    #model_or_pipeline="mskov/whisper-small.en",
    data=dataset,
    tokenizer=miso_tokenizer,
    #feature_extractor=miso_feature_extractor,
    input_column="audio",
    label_column="category",
    # device=None,
    strategy="simple",
    metric="wer",
)
print(results)


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

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