Pseudo-Cepstrum: Pitch Modification for Mel-Based Neural Vocoders

2025-12-18

AI Summary

Existing pitch modification methods for Mel-spectrum-driven neural vocoders rely either on fundamental frequency (F0) estimation or require model fine-tuning. To address these limitations, we propose a training-free, model-agnostic pitch shifting framework. Our approach first maps Mel spectrograms to a pseudo-cepstral domain via the pseudo-inverse Mel transform; then explicitly shifts harmonic peaks in the cepstral domain; and finally reconstructs the target Mel spectrogram through inverse discrete cosine transform (IDCT) and the Mel filterbank. This is the first method enabling universal, F0-estimation-free, and retraining-free pitch modification, implicitly modeling harmonic structure while overcoming accuracy and generalization bottlenecks inherent in conventional time-frequency domain approaches. Extensive evaluation on state-of-the-art vocoders—including HiFi-GAN and WaveGrad—demonstrates significant improvements in F0 root-mean-square error (RMSE) and Mel cepstral distortion (MCD) over PSOLA and WORLD, alongside marked gains in subjective Mean Opinion Score (MOS).

Technology Category

• Machine Learning: Mixture of Experts (MoE)
• Natural Language Processing: Speech
• Cognitive Modeling & Cognitive Systems: Neural Spike Coding

Application Category

• User Modeling, Personalization and Recommendation: User privacy protection in personalized systems
• Search and Retrieval-Augmented AI: Vertical and domain-specific search
• Security and Privacy: Privacy-enhancing technologies

Abstract

This paper introduces a cepstrum-based pitch modification method that can be applied to any mel-spectrogram representation. As a result, this method is compatible with any mel-based vocoder without requiring any additional training or changes to the model. This is achieved by directly modifying the cepstrum feature space in order to shift the harmonic structure to the desired target. The spectrogram magnitude is computed via the pseudo-inverse mel transform, then converted to the cepstrum by applying DCT. In this domain, the cepstral peak is shifted without having to estimate its position and the modified mel is recomputed by applying IDCT and mel-filterbank. These pitch-shifted mel-spectrogram features can be converted to speech with any compatible vocoder. The proposed method is validated experimentally with objective and subjective metrics on various state-of-the-art neural vocoders as well as in comparison with traditional pitch modification methods.

Problem

• Pitch modification for mel-based neural vocoders
• Compatible with any mel-based vocoder without retraining
• Shifts harmonic structure via cepstrum feature space

Innovation

• Cepstrum-based pitch modification for mel-spectrograms
• Compatible with any mel-based vocoder without retraining
• Direct harmonic structure shift via cepstral peak manipulation