音声の類似度を測定する方法を調べる

pip install transformers datasets torch

print(similarity)

tensor(0.9746, grad_fn=<SumBackward1>)

pip install librosa

from transformers import Wav2Vec2FeatureExtractor, WavLMForXVector
import torch
import librosa

model_name = "microsoft/wavlm-base-plus-sv"

feature_extractor = Wav2Vec2FeatureExtractor.from_pretrained(model_name)
model = WavLMForXVector.from_pretrained(model_name)


def get_audio_similarity(audio_path1, audio_path2) -> float:
    target_sr = 16000
    waveform1, sample_rate1 = librosa.load(audio_path1)
    waveform1 = librosa.resample(waveform1, orig_sr=sample_rate1, target_sr=target_sr)
    waveform2, sample_rate2 = librosa.load(audio_path2)
    waveform2 = librosa.resample(waveform2, orig_sr=sample_rate2, target_sr=target_sr)

    inputs = feature_extractor(
        [waveform1, waveform2],
        return_tensors="pt",
        padding=True,
    )

    embeddings = model(**inputs).embeddings
    embeddings = torch.nn.functional.normalize(embeddings, dim=-1).cpu()
    cosine_sim = torch.nn.CosineSimilarity(dim=-1)
    return cosine_sim(embeddings[0], embeddings[1]).item()


if __name__ == "__main__":
    audio_path1 = "speaker1_1.wav"
    audio_path2 = "speaker1_2.wav"
    similarity = get_audio_similarity(audio_path1, audio_path2)
    print(f"similarity : {similarity }")