Quranic Universal Aligner
المحاذي القرآني الشامل
Drop in a recitation, get word-level timing with confidence you can trust.
- stage
- Stable
- openness
- Contributions by invitation
The problem, solved once
Connecting a recording to its text is a need every reader, caption tool, and research pipeline shares. Each one has solved it alone, with its own model and its own failure modes.
The Aligner is one pipeline: voice activity detection, phoneme recognition, n-gram anchor voting, dynamic-programming alignment against the known text, and optional word, letter, and phoneme timing.
Confidence and failure are visible. Every segment carries its score, and the parts that did not align say so.
What you get
- PipelineVAD, phoneme ASR, anchor voting, DP alignment
- Phoneme ASRa 95M-parameter base model and a larger variant
- Optional word, letter, and phoneme timing through forced alignment
- Live on Hugging Face Spaces
Where we are, where we are heading
- nextLarger phoneme model
- nowVisible confidenceEvery segment carries its score; unaligned parts say so.
- 2026One pipeline, liveVAD, phoneme ASR, anchor voting, and DP alignment on Hugging Face Spaces.
How to help here
- Try a recitation and share a reproducible alignment issue with the maintainers.