What Do Forced Alignment Likelihood Scores Tell Us About the Aligned Speech
Bibliographic record
Abstract
Standard forced alignment systems are a widely used tool in phonetic research. Forced alignment uses Hidden Markov Models to align a sequence of phones to a sound recording. As a byproduct, it computes likelihood scores for every aligned phone and word. This study investigates the extent to which these likelihood scores can be: (1) pressed into use in speech research, (2) interpreted as a measure of acoustic distance (of some variety) to the modeled phones and place individual aligned segments within their distribution of phonetic variation. The present study is a first step in accomplishing these goals. To this end, first vowels in hold-out portions of the TIMIT (Zue & Seneff 1988) and Buckeye (Pitt et al. 2005) corpora were cross-aligned with phone models trained on the remaining portions of those corpora (tokens of [i] were aligned with the [?] phone, the [e] phone, the [?] phone, etc.), and the resulting likelihood scores were compared to acoustic measures like duration and formant frequencies to determine which acoustic properties are encapsulated in the scores. These were also compared with scores provided by the freely available Penn Forced Aligner (Yuan & Liberman, 2008). Preliminary analyses find a strong correlation between the cross alignment scores and F1 x F2 geometric distance, as well as the duration of the phones. This establishes that these probability measures show a relationship with some acoustic characteristics of the segments. The results of this initial analysis are promising. Future evaluation is needed to explore the full scope and limitations of the application of these measures. References: [1] Pitt, M. A., Johnson, K., Hume, E., Kiesling, S., & Raymond, W. 2005. The Buckeye corpus of conversational speech: labeling conventions and a test of transcriber reliability. Speech Communication 45, 89-95. [2] Zue, V. & Seneff, S. Transcription and Alignment of the TIMIT Database. Proceedings of the 2nd Meeting on Advanced Man -- Machine Interface through Spoken Language 1988, 11.1-11.10. [3] Yuan, J. & Liberman, M. 2008. Speaker identification on the SCOTUS corpus. Proceedings of Acoustics 2008.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.000 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.001 | 0.000 |
| Open science | 0.001 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.001 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one teacher head, not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".