app.py

import gradio as gr
import torch
from charts import spider_chart
from dictionaries import calculate_average, transform_dict
from icon import generate_icon
from transformers import pipeline
from timestamp import format_timestamp

MODEL_NAME = "openai/whisper-medium"
BATCH_SIZE = 8

device = 0 if torch.cuda.is_available() else "cpu"

pipe = pipeline(
    task="automatic-speech-recognition",
    model=MODEL_NAME,
    chunk_length_s=30,
    device=device,
)

#Define classifier for sentiment analysis
classifier = pipeline("text-classification", model="j-hartmann/emotion-english-distilroberta-base", top_k=None)

def transcribe(file, task, return_timestamps):
    outputs = pipe(file, batch_size=BATCH_SIZE, generate_kwargs={"task": task}, return_timestamps=True)
    text = outputs["text"]
    timestamps = outputs["chunks"]

    #If return timestamps is True, return html text with timestamps format
    if return_timestamps==True:
      spider_text = [f"{chunk['text']}" for chunk in timestamps] #Text for spider chart without timestamps
      timestamps = [f"[{format_timestamp(chunk['timestamp'][0])} -> {format_timestamp(chunk['timestamp'][1])}] {chunk['text']}" for chunk in timestamps]

    else:
      timestamps = [f"{chunk['text']}" for chunk in timestamps]
      spider_text = timestamps

    text = "<br>".join(str(feature) for feature in timestamps)
    text = f"<h4>Transcription</h4><div style='overflow-y: scroll; height: 250px;'>{text}</div>"

    spider_text = "\n".join(str(feature) for feature in spider_text)
    
    fig = spider_chart(calculate_average([transform_dict(classifier.predict(t)[0]) for t in spider_text.split("\n")]))

    return file, text, fig


inputs = [gr.Audio(source="upload", label="Audio file", type="filepath"),
        gr.Radio(["transcribe"], label="Task", value="transcribe"),
        gr.Checkbox(value=True, label="Return timestamps")]

outputs = [gr.Audio(label="Processed Audio", type="filepath"),
        gr.outputs.HTML("text"),
           gr.Plot(label="fig")]

title = "Whisper Demo: Transcribe Audio"

MODEL_NAME1 = "jpdiazpardo/whisper-tiny-metal"

description = ("Transcribe long-form audio inputs with the click of a button! Demo uses the"
        f" checkpoint [{MODEL_NAME1}](https://huggingface.co/{MODEL_NAME1}) and 🤗 Transformers to transcribe audio files"
        " of arbitrary length. Check some of the 'cool' examples below")

examples = [["When a Demon Defiles a Witch.wav","transcribe",True],
            ["Immaculate Misconception.wav","transcribe", True]]


linkedin = generate_icon("linkedin")
github = generate_icon("github")

article = ("<div style='text-align: center; max-width:800px; margin:10px auto;'>"
            f"<p>{linkedin} <a href='https://www.linkedin.com/in/juanpablodiazp/' target='_blank'>Juan Pablo Díaz Pardo</a><br>"
            f"{github} <a href='https://github.com/jpdiazpardo' target='_blank'>jpdiazpardo</a></p>"
          )


title = "Scream: Fine-Tuned Whisper model for automatic gutural speech recognition 🤟🤟🤟"

demo = gr.Interface(title = title, fn=transcribe, inputs = inputs, outputs = outputs, description=description, cache_examples=True, allow_flagging="never", article = article , examples=examples)

demo.queue(concurrency_count=3)
demo.launch(debug = True)