Neural Audio Fingerprint for High-Specific Audio Retrieval Based on Contrastive Learning

Chang, Sungkyun; Lee, Donmoon; Park, Jeongsoo; Lim, Hyungui; Lee, Kyogu; Ko, Karam; Han, Yoonchang

doi:10.1109/icassp39728.2021.9414337

Cited by 18 publications

(5 citation statements)

References 15 publications

(27 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Furthermore, our system performs a comprehensive search to precisely estimate the timestamp of the query in the identified reference audio using a simple sequence search strategy, which makes our system applicable to audio synchronization tasks. Our system performs well compared to the baseline [10] and Audfprint 4 methods. Also, our system is computationally and memoryefficient due to the compact embeddings that make our system deployable on an extensive database.…”

Section: Discussionmentioning

confidence: 91%

“…We compared our approach with a baseline system [10] that generates fingerprints using an encoder similar to Now-Playing's [9] architecture and does a comprehensive search. To the best of our knowledge, it is the only method that employs a neural network model to generate robust We trained the models with the Adam [21] optimizer for 150 epochs using the cyclic learning rate.…”

Section: Implementation Detailsmentioning

confidence: 99%

“…Google's music recognizer system [9] trains a CNN based on the triplet loss function to generate compact audio fin-gerprints. Another approach proposed in [10] trains a similar encoder as [9] using the contrastive learning framework and performs a maximum inner product search within a minibatch during training.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Attention-Based Audio Embeddings for Query-by-Example

Singh¹,

Demuynck²,

Arora³

2022

Preprint

View full text Add to dashboard Cite

An ideal audio retrieval system efficiently and robustly recognizes a short query snippet from an extensive database. However, the performance of well-known audio fingerprinting systems falls short at high signal distortion levels. This paper presents an audio retrieval system that generates noise and reverberation robust audio fingerprints using the contrastive learning framework. Using these fingerprints, the method performs a comprehensive search to identify the query audio and precisely estimate its timestamp in the reference audio. Our framework involves training a CNN to maximize the similarity between pairs of embeddings extracted from clean audio and its corresponding distorted and time-shifted version. We employ a channelwise spectral-temporal attention mechanism to better discriminate the audio by giving more weight to the salient spectral-temporal patches in the signal. Experimental results indicate that our system is efficient in computation and memory usage while being more accurate, particularly at higher distortion levels, than competing state-of-the-art systems and scalable to a larger database.

show abstract

Section: Discussionmentioning

confidence: 91%

Section: Implementation Detailsmentioning

confidence: 99%

See 1 more Smart Citation

Attention-Based Audio Embeddings for Query-by-Example

Singh¹,

Demuynck²,

Arora³

2022

Preprint

View full text Add to dashboard Cite

show abstract

“…Alternative systems with available implementations are by Chang et al (2021) and audfprint by Ellis (2014). Both systems however lack robustness to significant speed changes of more than 5%.…”

Section: Statement Of Needmentioning

confidence: 99%

Panako: a scalable audio search system

Six¹

2022

JOSS

View full text Add to dashboard Cite

Panako solves the problem of finding short audio fragments in large digital audio archives. The content based audio search algorithm implemented in Panako is able to identify a short audio query in a large database of thousands of hours of audio using an acoustic fingerprinting technique. Figure 1: A general acoustic fingerprinting system. Features are extracted from audio and combined into fingerprints. The fingerprints are matched with fingerprints in a reference database. Finally a match is reported.

show abstract

“…Especially in applications made with ANN (Artificial Neural Network), very high success rates are achieved. Studies numbered [20][21][22][23] are some important studies carried out in this field.…”

Section: Introductionmentioning

confidence: 99%

Effect of Spectrogram Parameters and Noise Types on The Performance of Spectro-temporal Peaks Based Audio Search Method

Köseoğlu

Uyanık

2023

Gazi University Journal of Science

View full text Add to dashboard Cite

Audio search algorithms are used to detect the matching file in large databases, especially in multimedia applications and smart appliances. Based on different audio fingerprint extraction methods, similar algorithms have been developed and applied in different fields. These algorithms are expected to perform the detection in a reliable and robust way within the possible shortest time. In this study, based on spectral peaks method, an audio fingerprint algorithm with a few minor modifications was developed to accurately detect the matching audio file in the target database. The algorithm was demonstrated and then the effect of spectrogram parameters such as window size, overlap and number of FFT was investigated in terms of reliability and robustness of the program under three different noise sources. The database was a relatively small size one with five genres of music. In this study, it was aimed to contribute to new audio file detection studies based on spectral peaks method. It was observed that the the variation in the spectrogram parameters have significantly affected the number of matchings (NM), reliability and robustness. Under high noise conditions the optimal spectrogram parameters were determined as 512;50%;512, respectively. We did not observe, however, a significant effect of music genre on NM.

show abstract

Neural Audio Fingerprint for High-Specific Audio Retrieval Based on Contrastive Learning

Cited by 18 publications

References 15 publications

Attention-Based Audio Embeddings for Query-by-Example

Attention-Based Audio Embeddings for Query-by-Example

Panako: a scalable audio search system

Effect of Spectrogram Parameters and Noise Types on The Performance of Spectro-temporal Peaks Based Audio Search Method

Contact Info

Product

Resources

About