Community-aware sampling improves text-to-speech training efficiency
Structure Before Sampling: Community-Aware Core-Set Selection for Data-Efficient Text-to-Speech
Computation and Language
Summary
Recording speech for computer voices is expensive, and many recorded sentences don’t add new sound information. The authors found that grouping similar sentences by their sounds reveals strong clusters. They created a method that picks sentences from different clusters and those with rare sounds to cover more sound variety efficiently. Training text-to-speech models on these smartly chosen sentences leads to better performance using much less data and time.
What this means in practice
- •For text-to-speech engineers: Train TTS models with subsets of utterances that represent varied sound patterns, reducing training data and time while improving accuracy.
- •For speech data collectors: Select speech samples from diverse phonetic groups to design more cost-efficient and varied TTS recording sessions.
Authors
Mizbaul Haque Maruf, Muhammad Nur Yanhaona
Abstract
Text-to-speech (TTS) corpora are costly to record, yet many utterances add little new phonetic information. Core-set selection reduces this cost by choosing a small training subset under a fixed audio-duration budget. We represent a corpus as a phonotactic graph that links each utterance to its most phonemically similar ones, and we first test whether this graph has structure. In Bangla and English corpora, its clustering is 199 and 56 times that of a size-matched random graph, and its modularity is more than twice that of a degree-preserving random graph. We then propose Community Representative, a selector that samples across graph communities and spreads its choices within each one, starting from utterances rich in rare phonemes. At every budget and in both languages, it covers more rare phoneme bigrams than random and entropy-based selection, and this lead holds on held-out utterances. TTS models trained on its 20% core-sets have a significantly lower character error rate (CER) than models trained on equal-duration random or entropy-based subsets in both languages. When all models train for the same number of epochs, the Bangla core-set model also outperforms full-corpus training (3.93% vs. 4.47% CER) with 4.5x less training time.