PRODIS -- a speech database and a phoneme-based language model for the study of predictability effects in Polish

TL;DR

PRODIS addresses the need for a large, high-quality Polish speech resource and a phoneme-level predictability metric. It combines a richly annotated corpus with a IPA-based GPT2-like architecture to estimate surprisal at the phoneme level, enabling robust analysis of how context influences acoustic distinctiveness. The processing pipeline automates transcription, alignment, and feature extraction, while the language component provides probabilistic predictions over IPA tokens to derive information-theoretic measures such as $Surprisal(unit_i) = -log_2 P(unit_i|Context)$. The dataset and methods are designed for extensibility to other Slavic languages, supporting advanced phonetic analysis and data-driven speech technologies.

Abstract

We present a speech database and a phoneme-level language model of Polish. The database and model are designed for the analysis of prosodic and discourse factors and their impact on acoustic parameters in interaction with predictability effects. The database is also the first large, publicly available Polish speech corpus of excellent acoustic quality that can be used for phonetic analysis and training of multi-speaker speech technology systems. The speech in the database is processed in a pipeline that achieves a 90% degree of automation. It incorporates state-of-the-art, freely available tools enabling database expansion or adaptation to additional languages.

Figures (1)

• Figure 1: PRODIS: database design (orange), speech processing (pink) and language modeling (blue) pipelines. The magnifying glass symbolises manual verification processes.