app.py

from difflib import Differ

import gradio as gr

import torch
from transformers import (
    AutoModelForSpeechSeq2Seq,
    AutoProcessor,
    pipeline,
)

description = """
 <div>
    <p>Roll up, roll up come test your diction against a 🤖</p>
 </div>
"""

diction_text = "How now brown cow"

test_text = f"""
    <div>
        <p>{diction_text}</p>
    </div>
"""

diction = gr.HTML(test_text)


device = "cpu"
# device = "cuda:0" if torch.cuda.is_available() else "cpu"
torch_dtype = torch.float16 if torch.cuda.is_available() else torch.float32

model_id = "openai/whisper-large-v3"


model = AutoModelForSpeechSeq2Seq.from_pretrained(
    model_id, low_cpu_mem_usage=True, use_safetensors=True
)
model.to(device)

processor = AutoProcessor.from_pretrained(model_id)

pipe = pipeline(
    task="automatic-speech-recognition",
    model=model,
    tokenizer=processor.tokenizer,
    feature_extractor=processor.feature_extractor,
    max_new_tokens=128,
    chunk_length_s=30,
    batch_size=8,
    return_timestamps=True,
    torch_dtype=torch_dtype,
    device=device,
)


def diff_texts(audio_input: str):
    test_text = diction_text
    d = Differ()

    return [
        (token[2:], token[0] if token[0] != "" else None)
        for token in d.compare(test_text, audio_input)
    ]


def transcribe_audio(diction, audio):
    print("Diction", diction)
    result = pipe(audio)
    print(f'TRANSCRIPTION {result["text"]}')
    diff_text = diff_texts(result["text"])

    print("diff", diff_text)

    return diff_text


highlighted_results = gr.HighlightedText(
    label="Diff",
    combine_adjacent=True,
    show_legend=True,
    color_map={"+": "red", "-": "green"},
)


input_audio = gr.Audio(
    sources=["microphone"],
    type="filepath",
    waveform_options=gr.WaveformOptions(
        waveform_color="#01C6FF",
        waveform_progress_color="#0066B4",
        skip_length=2,
        show_controls=False,
    ),
)

# demo = gr.Interface(
#     fn=transcribe_audio,
#     inputs=[diction, input_audio],
#     outputs=highlighted_results,
#     title="Test your diction",
#     description=description,
#     theme="abidlabs/Lime",
# )

with gr.Blocks() as demo:
    gr.HTML(description)
    gr.HTML(test_text)

    with gr.Row():
        inp = input_audio
        out = highlighted_results
    btn = gr.Button("Run")
    btn.click(fn=transcribe_audio, inputs=[diction_text, inp], outputs=out)


if __name__ == "__main__":
    demo.launch()