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 = """
    <div>
        <p>How now brown cow</p>
    </div>
"""

diction = gr.HTML(diction_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 transcribe_audio(audio):
    result = pipe(audio)
    print(f'TRANSCRIPTION {result["text"]}')
    try:
        for r in result:
            print(r)
    except:
        print("ERROR")
    return result


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="text",
    title="Test your diction",
    description=description,
    theme="abidlabs/Lime",
)

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