Less Peaky and More Accurate CTC Forced Alignment by Label Priors
This addresses forced alignment accuracy for speech processing applications, but is incremental as it builds on existing CTC methods with specific improvements.
The paper tackled the problem of CTC models producing peaky output distributions that cause inaccurate forced alignments, especially at phoneme level, by using label priors to boost paths with fewer blanks during training, resulting in 12-40% lower phoneme and word boundary errors on Buckeye and TIMIT datasets compared to standard CTC and heuristic methods.
Connectionist temporal classification (CTC) models are known to have peaky output distributions. Such behavior is not a problem for automatic speech recognition (ASR), but it can cause inaccurate forced alignments (FA), especially at finer granularity, e.g., phoneme level. This paper aims at alleviating the peaky behavior for CTC and improve its suitability for forced alignment generation, by leveraging label priors, so that the scores of alignment paths containing fewer blanks are boosted and maximized during training. As a result, our CTC model produces less peaky posteriors and is able to more accurately predict the offset of the tokens besides their onset. It outperforms the standard CTC model and a heuristics-based approach for obtaining CTC's token offset timestamps by 12-40% in phoneme and word boundary errors (PBE and WBE) measured on the Buckeye and TIMIT data. Compared with the most widely used FA toolkit Montreal Forced Aligner (MFA), our method performs similarly on PBE/WBE on Buckeye, yet falls behind MFA on TIMIT. Nevertheless, our method has a much simpler training pipeline and better runtime efficiency. Our training recipe and pretrained model are released in TorchAudio.