The speech forced-alignment line from Joseph Keshet's MLSpeech lab — two 2026 aligners that surpass the long-standing Montreal Forced Aligner (MFA).

Outputs 2

MWA (Multilingual Word-Level Forced Alignment)

model

<p>A SOTA word-level forced aligner (Interspeech 2026) that beats MFA on TIMIT and Buckeye and generalizes to the 1000+ languages of MMS; HF weights are adopted (~600+ downloads).</p>

Venue Interspeech 2026

FALCON

paper

<p>"FALCON: Fully Differentiable Neural Forced Alignment via Soft Dynamic Programming" &mdash; a differentiable soft-DP phoneme aligner that is end-to-end trainable and surpasses MFA across alignment thresholds.</p>

speechresearchopen-weight