v1.1.5

Author: tabahi

CHANGELOG.md

@@ -1,5 +1,13 @@
 # Changelog
 
+
+## [1.1.5] - 2026-06-05
+- Default `boundary_softness=3` is changed from 7 to 3 because at 3 it performs the best on TIMIT and buckeye dataset. 
+- Added warnings and parameter `self.warn_level` for verbose level. Use `PhonemeTimestampAligner.warn_level=3` for high verbosity.
+- Fixed small segments alignment. Previously one-word or half-words ("umm or oh") raised error of not enough phonemes for the given duration. That is patched up here: https://github.com/tabahi/bournemouth-forced-aligner/commit/e58757b5ff9c3cb71303067a8050b06d01f80097#diff-c530cc1a536c371d38e233b3192f5f7935f97342d6460e039edc8107fbdf0348
+
+
+
 ## [1.1.4] - 2026-02-28
 - `batch_size=16` parameter in `process_srt_file()` and `process_segments()`. Reduce it to prevent out-of-memory errors when processing large audio segments.
 
@@ -99,6 +107,8 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - click>=8.0.0
 - phonemizer>=3.3.0
 
+
+[1.1.5]: https://github.com/tabahi/bournemouth-forced-aligner/releases/tag/v1.1.5
 [1.1.3]: https://github.com/tabahi/bournemouth-forced-aligner/releases/tag/v1.1.3
 [1.1.2]: https://github.com/tabahi/bournemouth-forced-aligner/releases/tag/v1.1.2
 [1.1.0]: https://github.com/tabahi/bournemouth-forced-aligner/releases/tag/v1.1.0

README.md

@@ -623,7 +623,7 @@ PhonemeTimestampAligner(
     enforce_all_targets=True,     # Guarantee every phoneme in the transcript gets a timestamp
     ignore_noise=True,            # Skip predicted noise frames in output
     extend_soft_boundaries=True,  # Extend phoneme boundaries into adjacent low-confidence frames
-    boundary_softness=7,          # How far to extend (2=tight cores only, 7=generous)
+    boundary_softness=3,          # How far to extend (0=tight cores only, 7=generous)
     bad_confidence_threshold=0.6  # Flag segments where >60% of phonemes are low-confidence
 )
 ```
@@ -648,7 +648,7 @@ PhonemeTimestampAligner(
 | `ensure_completeness` | `False` | After decoding, inserts any missing phonemes at their best estimated position. Set `True` for extra completeness gaurantee. |
 | `ignore_noise` | `True` | Drops predicted noise/silence frames from output. Set `False` to include them as `"noise"` entries. |
 | `extend_soft_boundaries` | `True` | Extends phoneme boundaries into adjacent frames that still carry some acoustic evidence. |
-| `boundary_softness` | `7` | Controls how far boundaries extend. `2`–`3` = tight phoneme cores; `7` = generous boundaries. |
+| `boundary_softness` | `3` | Controls how far boundaries extend. `0`–`1` = tight phoneme cores; `7` = generous boundaries. `3` gives the best result on TIMIT|
 | `bad_confidence_threshold` | `0.6` | Ratio of low-confidence phonemes that triggers a `bad_alignment` warning on a segment. |
 
 **Model priority (highest → lowest):** `cupe_ckpt_path` → `model_name` → `preset` → defaults.
@@ -684,6 +684,9 @@ timestamps = aligner.process_srt_file(
     batch_size=4, # reduce if out-of-memory
     debug=True
 )
+
+
+aligner.warn_level = 3 # 0 = no warnings, 1 = important warnings (more critical for downstream tasks), 2 = all warnings (including low-confidence phonemes and bad confidence patterns). Critical errors will be raised as exceptions regardless of warn_level
 ```
 
 Returns a dict with a `"segments"` key. See [example_advanced.py](examples/example_advanced.py).
@@ -845,6 +848,8 @@ aligner.convert_to_textgrid(result, output_file="recording.TextGrid", include_co
 
 
 
+
+
 ---
 
 

pyproject.toml

@@ -5,7 +5,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "bournemouth-forced-aligner"
-version = "1.1.4"
+version = "1.1.5"
 description = "Bournemouth Forced Aligner - Phoneme-level timestamp extraction"
 readme = "README.md"
 requires-python = ">=3.8"

setup.py

@@ -12,7 +12,7 @@ def read_readme():
 
 setup(
     name="bournemouth-forced-aligner",
-    version="1.1.4",
+    version="1.1.5",
     packages=find_packages(exclude=["tests*", "docs*", "examples*"]),
     python_requires=">=3.8",