import gradio as gr
from transformers import pipeline
import numpy as np
"""
Globals!
"""
shortcut_js = """
"""
def fix_fmt(input_text):
# correct contractions (we cna't use the apostraphe)
input_text = input_text.replace("I'm", "I am")
input_text = input_text.replace("I'll", "I will")
input_text = input_text.replace("I'd", "I would")
input_text = input_text.replace("you'd", "you would")
input_text = input_text.replace("he'd", "he would")
input_text = input_text.replace("she'd", "she would")
input_text = input_text.replace("they'd", "they would")
input_text = input_text.replace("we'd", "we would")
input_text = input_text.replace("You'd", "You would")
input_text = input_text.replace("He'd", "He would")
input_text = input_text.replace("She'd", "She would")
input_text = input_text.replace("They'd", "They would")
input_text = input_text.replace("We'd", "We would")
input_text = input_text.replace("you'll", "you will")
input_text = input_text.replace("he'll", "he will")
input_text = input_text.replace("she'll", "she will")
input_text = input_text.replace("they'll", "they will")
input_text = input_text.replace("we'll", "we will")
input_text = input_text.replace("You'll", "You will")
input_text = input_text.replace("He'll", "He will")
input_text = input_text.replace("She'll", "She will")
input_text = input_text.replace("They'll", "They will")
input_text = input_text.replace("We'll", "We will")
# other contractions can be said without the apostraphe -- that is not an issue
input_text = input_text.replace("'", "")
input_text = input_text.replace('"', '')
return input_text
def get_html(input_text):
inner_html = """
Speech synthesiser
Speech System
"""
input_text = fix_fmt(input_text)
inner_html = inner_html.replace("SOME_DEFAULT_VALUE", input_text)
html = f"""
"""
return html
# this seems to have installed in a diff loc to where
# whisper.load_model(model_size) downlaods to
transcriber = pipeline(
"automatic-speech-recognition",
model="openai/whisper-medium" #large is also installed
)
def transcribe(stream, new_chunk):
try:
sr, y = new_chunk
y = y.astype(np.float32)
y /= np.max(np.abs(y))
if stream is not None:
stream = np.concatenate([stream, y])
else:
stream = y
text = transcriber(
{
"sampling_rate": sr,
"raw": stream
}
)["text"]
html = get_html(text)
return stream, html
except:
return None, None
def run_demo():
demo = gr.Interface(
transcribe,
inputs = [
"state",
gr.Audio(sources=["microphone"], streaming=False)
],
outputs = [
"state",
gr.HTML()
],
live=True,
head=shortcut_js
)
demo.launch()
if __name__ == "__main__":
run_demo()