List your company

ResearchFormerly WizWand

Forced Alignment

Benchmarks

DatasetSOTA methodMetricTrendResultsLast updated

MFA-Labeled Raw (test)

Qwen3-ForcedAligner-0.6B

42.9AAS Latency (Avg)

8

Feb 26, 2026

LibriSpeech Other

96.64AAS

5

Jun 1, 2026

LibriSpeech Clean

NeMo Forced Aligner

87.05AAS

5

Jun 1, 2026

GTSinger-Speech-ZH

221.29AAS

5

Jun 1, 2026

MFA-labeled Long-form (test)

LLM-ForcedAligner

52.9Average Alignment Value

4

Feb 26, 2026

Human-Labeled (test)

0.007Avg. RTF

4

Feb 26, 2026

MFA-Labeled Concat-300s (test)

Qwen3-ForcedAligner-0.6B

52.9AAS (Avg) [ms]

4

Feb 26, 2026

human-labeled Chinese datasets (Mixed-300s)

Monotonic-Aligner

410.8AAS

3

Feb 26, 2026

human-labeled Chinese datasets (Mixed-60s)

86.7AAS

3

Feb 26, 2026

human-labeled Chinese datasets (Raw-Noisy)

0.895AAS

3

Feb 26, 2026

human-labeled Chinese datasets (Raw)

88.6AAS

3

Feb 26, 2026

15.9Mean Absolute Error (ms)

2

Apr 15, 2026

13.39Mean Absolute Error (ms)

2

Apr 15, 2026

13.64Mean Abs Error (ms)

2

Apr 15, 2026

Randomly selected audio files and transcriptions Manual Inspection

81AU Count

2

Feb 26, 2026

human-labeled Chinese and MFA-labeled multilingual speech Mixed-Crosslingual

LLM-ForcedAligner

42.5AAS

1

Feb 26, 2026