Common Voice Scripted Speech 26.0 - American English (Male)

Splits and speakers: 295,743 validated clips total — train 280,865 clips / 5,405 speakers, dev 7,393 clips / 156 speakers, test 7,485 clips / 144 speakers. Split is speaker-disjoint (no client_id appears in more than one split), proportioned 95.0/2.5/2.5 by clip count via a greedy per-speaker balancing algorithm (speakers have uneven clip counts, so the split is a close but not exact match to the 95/2.5/2.5 target while keeping speakers disjoint).

Clip duration: 390.03 hours total (1,404,097 seconds) across 295,743 clips. Per-split: train 370.60h, dev 10.07h, test 9.36h. Per-clip: mean 4.75s, median 4.65s, min 0.74s, max 26.74s, std 1.64s.

Clip duration histogram:

   0-1s |                                                         20 ( 0.0%)
   1-2s | #####                                                6,866 ( 2.3%)
   2-3s | ############################                        37,687 (12.7%)
   3-4s | #############################################       60,632 (20.5%)
   4-5s | ##################################################  66,605 (22.5%)
   5-6s | ############################################        59,030 (20.0%)
   6-7s | ############################                        37,484 (12.7%)
   7-8s | #############                                       17,359 ( 5.9%)
   8-9s | #####                                                6,831 ( 2.3%)
  9-10s | #                                                    2,607 ( 0.9%)
 10-12s |                                                        570 ( 0.2%)
 12-15s |                                                         35 ( 0.0%)
 15-20s |                                                          9 ( 0.0%)
 20-30s |                                                          8 ( 0.0%)
   30s+ |                                                          0 ( 0.0%)

Gender distribution: 100% male_masculine (295,743 clips).

Age distribution (shipped subset, self-declared):

Accent tags: closed-form matching means the accents field is exactly "United States English" for every shipped clip — there is no multi-tag distribution to report (unlike the British-English submission's open-form table, where 5.9% of clips carried a co-occurring non-UK tag). QA confirmation: 100% of the 295,743 shipped clips have exactly one accent tag, verified programmatically before packaging.

Open-form vs. closed-form filtering comparison (checked against live metadata, male_masculine gender only, since this is the deliberate divergence point from the British-English reference pipeline):

Full-corpus context: the full validated English corpus has 1,901,978 rows; gender raw values are male_masculine 891,316 / (blank) 697,742 / female_feminine 312,710 / do_not_wish_to_say 124 / non-binary 69 / transgender 16 / intersex 1. The variant column is entirely blank in this release (no separate "variety" dimension). Full accent x gender and age x gender cross-tabs are in the standalone stats.md.

Derivation methodology: derived from the Mozilla Common Voice Scripted Speech 26.0 English release (cv-corpus-26.0-2026-06-12, locale en). Filter: validated rows only (present in the source validated.tsv), whose self-declared accents field is exactly "United States English" (closed-form, no co-occurring tags — re-verified against the live metadata distribution, not assumed from a prior snapshot), plus gender == "male_masculine". No age filter applied. Speaker-disjoint train/dev/test split via a greedy per-speaker assignment balancing toward 95/2.5/2.5 targets. Only the referenced MP3 clips were extracted from the ~94.6GB source archive (not the full archive) via a single sequential streaming pass over the source tar.gz.