app.py

import gradio as gr
import torch
import torchaudio
from torchaudio.transforms import Resample
from transformers import AutoFeatureExtractor, AutoModelForAudioXVector

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

STYLE = """
<link rel="stylesheet" href="https://cdn.jsdelivr.net/npm/[email protected]/dist/css/bootstrap.min.css" integrity="sha256-YvdLHPgkqJ8DVUxjjnGVlMMJtNimJ6dYkowFFvp4kKs=" crossorigin="anonymous">
"""
OUTPUT_OK = (
    STYLE
    + """
    <div class="container">
        <div class="row"><h1 style="text-align: center">The speakers are</h1></div>
        <div class="row"><h1 class="display-1 text-success" style="text-align: center">{:.1f}%</h1></div>
        <div class="row"><h1 style="text-align: center">similar</h1></div>
        <div class="row"><h1 class="text-success" style="text-align: center">Welcome, human!</h1></div>
        <div class="row"><small style="text-align: center">(You must get at least 80% to be considered the same person)</small><div class="row">
    </div>
"""
)
OUTPUT_FAIL = (
    STYLE
    + """
    <div class="container">
        <div class="row"><h1 style="text-align: center">The speakers are</h1></div>
        <div class="row"><h1 class="display-1 text-danger" style="text-align: center">{:.1f}%</h1></div>
        <div class="row"><h1 style="text-align: center">similar</h1></div>
        <div class="row"><h1 class="text-danger" style="text-align: center">You shall not pass!</h1></div>
        <div class="row"><small style="text-align: center">(You must get at least 80% to be considered the same person)</small><div class="row">
    </div>
"""
)

THRESHOLD = 0.80

model_name = "microsoft/wavlm-base-plus-sv"
feature_extractor = AutoFeatureExtractor.from_pretrained(model_name)
model = AutoModelForAudioXVector.from_pretrained(model_name).to(device)
cosine_sim = torch.nn.CosineSimilarity(dim=-1)


def preprocess_audio(file_path, target_sr=16000):
    wav, sr = torchaudio.load(file_path)
    if sr != target_sr:
        wav = Resample(orig_freq=sr, new_freq=target_sr)(wav)
    return wav


def similarity_fn(path1, path2):
    if not (path1 and path2):
        return '<b style="color:red">ERROR: Please record audio for *both* speakers!</b>'

    wav1 = preprocess_audio(path1)
    wav2 = preprocess_audio(path2)

    input1 = feature_extractor(wav1.squeeze(0), return_tensors="pt", sampling_rate=16000).input_values.to(device)
    input2 = feature_extractor(wav2.squeeze(0), return_tensors="pt", sampling_rate=16000).input_values.to(device)

    with torch.no_grad():
        emb1 = model(input1).embeddings
        emb2 = model(input2).embeddings
    emb1 = torch.nn.functional.normalize(emb1, dim=-1).cpu()
    emb2 = torch.nn.functional.normalize(emb2, dim=-1).cpu()
    similarity = cosine_sim(emb1, emb2).numpy()[0]

    if similarity >= THRESHOLD:
        output = OUTPUT_OK.format(similarity * 100)
    else:
        output = OUTPUT_FAIL.format(similarity * 100)

    return output


with gr.Blocks() as demo:
    gr.Markdown("# Voice Authentication with WavLM + X-Vectors")
    gr.Markdown(
        "This demo compares two speech samples to determine if they are from the same speaker. "
        "Try it with your own voice!"
    )

    with gr.Row():
        input1 = gr.Audio(sources=["microphone", "upload"], type="filepath", label="Speaker #1")
        input2 = gr.Audio(sources=["microphone", "upload"], type="filepath", label="Speaker #2")

    output = gr.HTML(label="Result")

    btn = gr.Button("Compare Speakers")
    btn.click(similarity_fn, inputs=[input1, input2], outputs=output)

    gr.Examples(
        examples=[
            ["samples/denzel_washington.mp3", "samples/denzel_washington.mp3"],
            ["samples/heath_ledger_2.mp3", "samples/heath_ledger_3.mp3"],
            ["samples/heath_ledger_3.mp3", "samples/denzel_washington.mp3"],
            ["samples/denzel_washington.mp3", "samples/heath_ledger_2.mp3"],
        ],
        inputs=[input1, input2],
    )

    gr.Markdown(
        "<p style='text-align: center'>"
        "<a href='https://huggingface.co/microsoft/wavlm-base-plus-sv' target='_blank'>🎙️ Learn more about WavLM</a> | "
        "<a href='https://arxiv.org/abs/2110.13900' target='_blank'>📚 WavLM paper</a> | "
        "<a href='https://www.danielpovey.com/files/2018_icassp_xvectors.pdf' target='_blank'>📚 X-Vector paper</a>"
        "</p>"
    )

demo.launch()