Towards confidence measures on fundamental frequency estimations

Deng, Boyuan; Jouvet, Denis; Laprie, Yves; Steiner, Ingmar; Sini, Aghilas

doi:10.1109/icassp.2017.7953229

Cited by 6 publications

(4 citation statements)

References 10 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…However, the confidence of the estimated pitch in the voiced case is seldom provided. A few exceptions are CREPE [14], which produces a confidence score computed from the activations of the last layer of the model, and [25], which directly addresses this problem, by training a neural network based on hand-crafted features to estimate the confidence of the estimated pitch. In contrast, in our work we explicitly augment the proposed model with a head aimed at estimating confidence in a fully unsupervised way.…”

Section: Related Workmentioning

confidence: 99%

SPICE: Self-Supervised Pitch Estimation

Gfeller

Frank

Roblek

et al. 2020

IEEE/ACM Trans. Audio Speech Lang. Process.

View full text Add to dashboard Cite

Section: Related Workmentioning

confidence: 99%

SPICE: Self-Supervised Pitch Estimation

Gfeller

Frank

Roblek

et al. 2020

IEEE/ACM Trans. Audio Speech Lang. Process.

View full text Add to dashboard Cite

“…Currently there is no indication of the reliability of the estimated F0 values provided by the various pitch detection algorithms. Some preliminary work has been carried out in this direction [15], but further studies are still necessary.…”

Section: Fundamental Frequencymentioning

confidence: 99%

Speech Processing and Prosody

Jouvet

2019

Text, Speech, and Dialogue

Self Cite

View full text Add to dashboard Cite

The prosody of the speech signal conveys information over the linguistic content of the message: prosody structures the utterance, and also brings information on speaker's attitude and speaker's emotion. Duration of sounds, energy and fundamental frequency are the prosodic features. However their automatic computation and usage are not obvious. Sound duration features are usually extracted from speech recognition results or from a force speech-text alignment. Although the resulting segmentation is usually acceptable on clean native speech data, performance degrades on noisy or not non-native speech. Many algorithms have been developed for computing the fundamental frequency, they lead to rather good performance on clean speech, but again, performance degrades in noisy conditions. However, in some applications, as for example in computer assisted language learning, the relevance of the prosodic features is critical; indeed, the quality of the diagnostic on the learner's pronunciation will heavily depend on the precision and reliability of the estimated prosodic parameters. The paper considers the computation of prosodic features, shows the limitations of automatic approaches, and discusses the problem of computing confidence measures on such features. Then the paper discusses the role of prosodic features and how they can be handled for automatic processing in some tasks such as the detection of discourse particles, the characterization of emotions, the classification of sentence modalities, as well as in computer assisted language learning and in expressive speech synthesis.

show abstract

“…However, the confidence of the estimated pitch in the voiced case is seldom provided. A few exceptions are CREPE [12], which produces a confidence score computed from the activations of the last layer of the model, and [22], which directly addresses this problem, by training a neural network based on handcrafted features to estimate the confidence of the estimated pitch. In contrast, in our work we explicitly augment the proposed model with a head aimed at estimating confidence in a fully unsupervised way.…”

Section: Related Workmentioning

confidence: 99%

SPICE: Self-supervised Pitch Estimation

Gfeller,

Frank,

Roblek

et al. 2019

Preprint

View full text Add to dashboard Cite

We propose a model to estimate the fundamental frequency in monophonic audio, often referred to as pitch estimation. We acknowledge the fact that obtaining ground truth annotations at the required temporal and frequency resolution is a particularly daunting task. Therefore, we propose to adopt a self-supervised learning technique, which is able to estimate (relative) pitch without any form of supervision. The key observation is that pitch shift maps to a simple translation when the audio signal is analysed through the lens of the constant-Q transform (CQT). We design a self-supervised task by feeding two shifted slices of the CQT to the same convolutional encoder, and require that the difference in the outputs is proportional to the corresponding difference in pitch. In addition, we introduce a small model head on top of the encoder, which is able to determine the confidence of the pitch estimate, so as to distinguish between voiced and unvoiced audio. Our results show that the proposed method is able to estimate pitch at a level of accuracy comparable to fully supervised models, both on clean and noisy audio samples, yet it does not require access to large labeled datasets.

show abstract

Towards confidence measures on fundamental frequency estimations

Cited by 6 publications

References 10 publications

SPICE: Self-Supervised Pitch Estimation

SPICE: Self-Supervised Pitch Estimation

Speech Processing and Prosody

SPICE: Self-supervised Pitch Estimation

Contact Info

Product

Resources

About