We present a blank-aware decoding approach for transcript-free phoneme alignment with CTC-based speech foundation models, designed to improve annotation bootstrapping in low-resource languages. While CTC models provide frame-level phoneme posteriors without requiring transcripts, greedy decoding produces blank-dominated and temporally unstable segmentations that are difficult to correct manually. Our approach introduces two training-free blank-resolution strategies operating directly on CTC logits: (i) confidence-ratio substitution, which promotes competitive non-blank hypotheses relative to the blank symbol, and (ii) recursive context adjustment, which enforces local contextual consistency within blank spans. Experiments on English (TIMIT) and on Sardinian and Tyrolean dialect corpora show consistent improvements in boundary F1 prediction, phoneme duration regularity, and segmentation stability over greedy CTC decoding. Although absolute boundary deviations remain higher than transcript-conditioned aligners, the resulting alignments are structurally coherent and suitable for manual correction. A post-hoc phoneme-class analysis further reveals systematic asymmetries in blank resolution, highlighting complementary roles of local acoustic evidence and contextual cues, and outlining prominising venues for future improvements.

We present WikIPA, a new multilingual benchmark designed for automatic speech-to-IPA (STIPA) transcription. By integrating human-curated IPA transcriptions from WikiPron with spoken recordings and metadata from Lingua Libre, WikIPA connects textual phonetic representations with real speech across 78 languages. This open resource supports both broad (phonemic) and narrow (phonetic) transcription tasks, enabling fine-grained evaluation of multilingual phonetic transcription systems. WikIPA provides over 289,000 paired entries and serves as a large-scale foundation for STIPA. We benchmark several state-of-the-art STIPA systems, including MultIPA, (Lo)WhIPA, and ZIPA. Results show that ZIPA achieves the lowest mean error rates across most languages, outperforming Whisper- and Wav2Vec-based baselines. Error analyses reveal that remaining discrepancies largely stem from minor phonetic confusions rather than complete transcription failures, emphasizing the challenge of modeling fine-grained articulatory variation. WikIPA thus establishes the first systematic, multilingual evaluation framework for speech-to-IPA transcription and highlights the potential of combining open, community-driven resources to advance STIPA evaluation.

Automatic Speech Recognition systems (ASR) based on neural networks achieve great results, but it remains unclear which are the linguistic features and representations that the models leverage to perform the recognition. In our study, we used phonological syllables as tokens to fine-tune an end-to-end ASR model due to their relevance as linguistic units. Furthermore, this strategy allowed us to keep track of different types of linguistic features characterizing the tokens. The analysis of the transcriptions generated by the model reveals that factors such as token frequency and lexical stress have a variable impact on the prediction strategies adopted by the ASR system.