How kb-labb/easyaligner works
easyaligner is a Python forced-alignment library: users provide audio, transcript segments, and an acoustic model; the pipeline detects speech regions, extracts model emissions, then aligns normalized transcript tokens to audio and writes timestamped results. Text normalization preserves links to the original transcript, while tokenization and fuzzy matching support flexible segmentation and transcript matching. The README documents separate stage outputs and both known- and unknown-region workflows. The graph reflects those documented stages and indexed module references; several internal calls and optional paths were not shown in the excerpts.
Main entry points: pipelines.py Pipeline orchestration · pytorch.py Forced alignment · datamodel.py Speech data model · normalization.py Text normalization
Components 22
Alignment workflow
- Pipeline orchestration
src/easyaligner/pipelines.py - VAD stage
src/easyaligner/pipelines.py - Emission extraction
src/easyaligner/pipelines.py - Forced alignment
src/easyaligner/alignment/pytorch.py - Alignment results
src/easyaligner/data/datamodel.py - Alignment utilities
src/easyaligner/alignment/utils.py
Text processing
- Text normalization
src/easyaligner/text/normalization.py - Text tokenization
src/easyaligner/text/tokenizer.py - Swedish rules
src/easyaligner/text/languages/sv.py - Fuzzy matching
src/easyaligner/text/match.py
Audio and metadata
- Speech data model
src/easyaligner/data/datamodel.py - Audio datasets
src/easyaligner/data/dataset.py - Batch collators
src/easyaligner/data/collators.py - Audio data utilities
src/easyaligner/data/utils.py - Stage file output
src/easyaligner/utils.py
Speech detection
- VAD selection
src/easyaligner/vad/vad.py - Pyannote VAD
src/easyaligner/vad/pyannote.py - Silero VAD
src/easyaligner/vad/silero.py - VAD utilities
src/easyaligner/vad/utils.py
Other components
- Library user
- CTC acoustic model
- Output files
Connections 23
- Library user → Pipeline orchestration: runs
README.md - Pipeline orchestration → VAD stage: runs VAD
src/easyaligner/pipelines.py - Pipeline orchestration → Emission extraction: extracts emissions
README.md - Pipeline orchestration → Forced alignment: runs alignment
README.md - VAD stage → Speech data model: records segments
src/easyaligner/pipelines.py - Emission extraction → CTC acoustic model: infers logits
src/easyaligner/pipelines.py - Forced alignment → Text normalization: normalizes tokens
src/easyaligner/alignment/pytorch.py - Forced alignment → Alignment utilities: uses utilities
src/easyaligner/alignment/pytorch.py - Forced alignment → Speech data model: reads metadata
src/easyaligner/alignment/pytorch.py - Forced alignment → Alignment results: produces alignments
README.md - Text normalization → Swedish rules: uses language rules
src/easyaligner/text/normalization.py - Fuzzy matching → Text normalization: uses normalization
src/easyaligner/text/match.py - Audio datasets → Speech data model: loads metadata
src/easyaligner/data/dataset.py - Pipeline orchestration → Audio datasets: uses datasets
src/easyaligner/pipelines.py - Pipeline orchestration → Batch collators: batches features
src/easyaligner/pipelines.py - Pipeline orchestration → Stage file output: saves stage files
src/easyaligner/pipelines.py - Stage file output → Output files: writes files
src/easyaligner/utils.py - VAD selection → Pyannote VAD: selects backend
src/easyaligner/vad/vad.py - VAD selection → Silero VAD: selects backend
src/easyaligner/vad/vad.py - Pyannote VAD → VAD utilities: uses helpers
src/easyaligner/vad/pyannote.py - Silero VAD → VAD utilities: uses helpers
src/easyaligner/vad/silero.py - VAD utilities → Speech data model: builds segments
src/easyaligner/vad/utils.py - Text tokenization → Library user: segments transcript