Datasets:
VoxCPM2 first-patch listening check (#31)
Listening pairs for issue #31 (first audio under 100 ms p99 at 200 requests at once): the same sentence generated twice, both in waves of 128 new requests. main solves the first 160 ms patch in 3 LocDiT steps; branch #31 solves it in 2 (the grid of the 2-step student the 3-step one was trained from), which cuts ~11 ms of GPU from the first audio at 200 requests. Every later patch is solved in 3 steps in both, but the rest of the sentence follows the first patch, so the two versions are different takes of the sentence, not the same audio with a different start.
On the #2 set (1,320 clips) nothing changed significantly over whole clips (CER, WER, UTMOS, DNSMOS, speaker similarity). The first second's UTMOS is lower: -0.04 (2.46 -> 2.42, 95 % interval -0.065 to -0.022). Listen to the start of each pair: if branch #31's openings sound worse (noisier, muffled, a click or a smeared first syllable), the change should not ship.
The dataset viewer above plays both versions of a sentence side by side: one row per dialect, the main and branch_31 columns.
Each clip is the opening of a sentence, cut after whole words (at least 3 words and 1 s), checked with Whisper large-v3 so that the words are heard.
| dialect | words | length | main | branch #31 |
|---|---|---|---|---|
| البحرينية (bh) | والعكس أيضا، الشركات | 2.4 s | main | branch #31 |
| الجزائرية (dz) | في جنينت القصر | 1.7 s | main | branch #31 |
| الجزائرية (dz) | وٱلحمد لله، بفضل | 2.5 s | main | branch #31 |
| الجزائرية (dz) | دخلو للمدينة، صرفو | 2.0 s | main | branch #31 |
| المصرية (eg) | بأعتقد إن الماسونية | 1.4 s | main | branch #31 |
| الحجازية (hejazi) | طيب خلاص الصبح | 2.2 s | main | branch #31 |
| العراقية (iq) | هو يشتغل صباغ، | 1.7 s | main | branch #31 |
| الخليجية (khaliji) | شفتي شلون بيتنا | 1.9 s | main | branch #31 |
| الخليجية (khaliji) | عبوتي عاد عبوتي | 1.4 s | main | branch #31 |
| الخليجية (khaliji) | واليوم الفجر طلع | 1.9 s | main | branch #31 |
| اللبنانية (lb) | فنجان مشقوق، بضل | 2.4 s | main | branch #31 |
| الليبية (ly) | وبعدين لما تسأل | 1.2 s | main | branch #31 |
| المغربية (ma) | ودبا خاصكم تديوها | 2.0 s | main | branch #31 |
| الفصحى (msa) | يقال أن ثقافة | 1.3 s | main | branch #31 |
| النجدية (najdi) | القرية هذه مهجورة | 2.0 s | main | branch #31 |
| النجدية (najdi) | بس ان شاء الله | 1.0 s | main | branch #31 |
| النجدية (najdi) | يعني أنا أفهم | 1.3 s | main | branch #31 |
| الفلسطينية (ps) | فلما هي تشعر | 1.6 s | main | branch #31 |
| السعودية (sa) | يوصل لكم في أي مكان، | 1.2 s | main | branch #31 |
| السودانية (sd) | ال اللي هو | 1.5 s | main | branch #31 |
| السورية (sy) | بوجود القيادات القوية | 1.7 s | main | branch #31 |
| التونسية (tn) | ٱلإنسان المقابل، المريض | 2.3 s | main | branch #31 |
| التونسية (tn) | ٱلميل ٱلجنسي موجود | 1.4 s | main | branch #31 |
| اليمنية (ye) | لن يحدث هذا | 1.5 s | main | branch #31 |
Skipped: 50 sentence(s) that failed the word check (no word heard at the start, an opening that does not match the text's first word, or a cut clip that lost its words): dz-0039_s2000, hejazi-0422_s2000, iq-0307_s2000, iq-0302_s2000, iq-0308_s2000, iq-0329_s2000, iq-0309_s2000, khaliji-0559_s2000, khaliji-0451_s2000, khaliji-0474_s2000, khaliji-0473_s2000, khaliji-0558_s2000, khaliji-0467_s2000, khaliji-0556_s2000, lb-0111_s2000, ly-0127_s2000, ly-0135_s2000, ly-0125_s2000, ma-0349_s2000, msa-0494_s2000, sa-0384_s2000, sd-0623_s2000, sd-0184_s2000, sy-0630_s2000, tn-0242_s2000, tn-0241_s2000, ye-0295_s2000, khaliji-0450_s2000, khaliji-0477_s2000, khaliji-0462_s2000, khaliji-0471_s2000, khaliji-0479_s2000, khaliji-0465_s2000, najdi-0399_s2000, najdi-0596_s2000, dz-0030_s2000, dz-0057_s2000, dz-0053_s2000, dz-0520_s2000, dz-0046_s2000, dz-0043_s2000, dz-0033_s2000, dz-0047_s2000, tn-0258_s2000, tn-0246_s2000, tn-0649_s2000, tn-0256_s2000, tn-0265_s2000, tn-0643_s2000, tn-0269_s2000.
Gated: the voices are cloned from the gated Arabic-Hindi-Reference-Voices.
- Downloads last month
- 21