# Phonetic Error Analysis of Raw Waveform Acoustic Models

arXiv: https://arxiv.org/abs/2606.07030 · HTML: https://mlsanity.com/paper/2606.07030/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 10 of its 43 references that are in the map.
- First posted on arXiv 5 Jun 2026.

## Builds on (most cited first, 10 of 10)

- [Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift](https://mlsanity.com/paper/1502.03167/) (2015)
- [Phase-aware Speech Enhancement with Deep Complex U-Net](https://mlsanity.com/paper/1903.03107/) (2019)
- [The PyTorch-Kaldi Speech Recognition Toolkit](https://mlsanity.com/paper/1811.07453/) (2018)
- [Learning Multiscale Features Directly From Waveforms](https://mlsanity.com/paper/1603.09509/) (2016)
- [CGCNN: Complex Gabor Convolutional Neural Network on raw speech](https://mlsanity.com/paper/2002.04569/) (2020)
- [The Role of Phonetic Units in Speech Emotion Recognition](https://mlsanity.com/paper/2108.01132/) (2021)
- [Incorporating Broad Phonetic Information for Speech Enhancement](https://mlsanity.com/paper/2008.07618/) (2020)
- [Learning Waveform-Based Acoustic Models using Deep Variational Convolutional Neural Networks](https://mlsanity.com/paper/1906.09526/) (2019)
- [Multi-Span Acoustic Modelling using Raw Waveform Signals](https://mlsanity.com/paper/1906.11047/) (2019)
- [Acoustic Model Adaptation from Raw Waveforms with SincNet](https://mlsanity.com/paper/1909.13759/) (2019)

## Abstract (arXiv)

We analyse error patterns of raw waveform acoustic models on TIMIT phone recognition beyond the overall phone error rate (PER). PER is decomposed across three broad phonetic class (BPC) categorisations, and confusion matrices are constructed from substitution errors. Our models combine parametric (SincNet, Sinc2Net) or non-parametric CNNs with Bidirectional LSTMs, achieving 13.9%/15.3% PER on Dev/Test, the best reported results for raw waveform models on TIMIT. Transfer learning from WSJ reduces PER to 11.3%/12.3%, surpassing the Filterbank baseline. Per-BPC analysis reveals that BLSTM layers benefit transition-dependent classes most, while WSJ transfer learning improves consonants roughly three times more than vowels. Confusion patterns are consistent across raw waveform and Filterbank systems, indicating that the dominant confusions reflect inherent phonetic similarities.