Datasets:

KaniTTS-research-team
/

emolia_filter_sr_32

category stringclasses 4 values	bad_prob float64 0.01 0.97	sample_id stringlengths 8 8	speaker_status stringclasses 3 values	speaker_overlap float64 0.01 0.95	sampling_rate int64 32k 32k
best_good	0.007	8108a004	UNCERTAIN	0.3522	32,000
best_good	0.009	057f73b2	GOOD	0.083	32,000
best_good	0.013	a77814d8	GOOD	0.1122	32,000
best_good	0.017	bbe3019f	GOOD	0.1505	32,000
best_good	0.018	a4f53bc4	GOOD	0.2385	32,000
best_good	0.019	33053322	BAD	0.9548	32,000
best_good	0.02	f058ed03	GOOD	0.0438	32,000
best_good	0.022	3a0f4409	GOOD	0.1131	32,000
best_good	0.026	1731970b	UNCERTAIN	0.3616	32,000
best_good	0.027	72973bef	GOOD	0.0269	32,000
best_good	0.027	9639859b	GOOD	0.0876	32,000
best_good	0.027	96413ac6	UNCERTAIN	0.3778	32,000
best_good	0.027	986f8e44	GOOD	0.0903	32,000
best_good	0.028	f8721e66	GOOD	0.0147	32,000
best_good	0.029	a76f4ae4	BAD	0.8377	32,000
boundary_good	0.347	c3f5e742	GOOD	0.0126	32,000
boundary_good	0.347	c4b42658	GOOD	0.0694	32,000
boundary_good	0.348	1a160cde	GOOD	0.0258	32,000
boundary_good	0.348	431fdf98	GOOD	0.0538	32,000
boundary_good	0.348	4e45a659	GOOD	0.0256	32,000
boundary_good	0.348	8184c7f5	GOOD	0.2154	32,000
boundary_good	0.348	d3fc357b	GOOD	0.0327	32,000
boundary_good	0.348	e0be0378	GOOD	0.0484	32,000
boundary_good	0.349	053d88d6	GOOD	0.0197	32,000
boundary_good	0.349	05c56879	GOOD	0.2843	32,000
boundary_good	0.349	24310d74	UNCERTAIN	0.3375	32,000
boundary_good	0.349	4a504e30	GOOD	0.1939	32,000
boundary_good	0.349	5f5e92e3	GOOD	0.0081	32,000
boundary_good	0.349	b9465dd8	GOOD	0.0853	32,000
boundary_good	0.35	a23a65e0	GOOD	0.0143	32,000
boundary_bad	0.35	968d3619	GOOD	0.0865	32,000
boundary_bad	0.351	1a9f84ae	GOOD	0.0187	32,000
boundary_bad	0.351	2ed22649	GOOD	0.0642	32,000
boundary_bad	0.351	d7ff1b02	GOOD	0.0713	32,000
boundary_bad	0.351	eb2878ea	GOOD	0.0284	32,000
boundary_bad	0.352	2d437d82	GOOD	0.0158	32,000
boundary_bad	0.352	45b44d0e	UNCERTAIN	0.2032	32,000
boundary_bad	0.352	6eb8bdd3	GOOD	0.0522	32,000
boundary_bad	0.352	a59750ad	GOOD	0.063	32,000
boundary_bad	0.352	b0813b0e	GOOD	0.0629	32,000
boundary_bad	0.352	fa1e7aa8	GOOD	0.1687	32,000
boundary_bad	0.353	0a1dccf7	GOOD	0.0742	32,000
boundary_bad	0.353	c5bb46bc	GOOD	0.0966	32,000
boundary_bad	0.353	c657ed96	GOOD	0.1791	32,000
boundary_bad	0.353	fffefac8	GOOD	0.1507	32,000
worst_bad	0.963	2f4d60b4	UNCERTAIN	0.2353	32,000
worst_bad	0.964	082b4117	GOOD	0.049	32,000
worst_bad	0.964	bccd8413	BAD	0.3505	32,000
worst_bad	0.964	cc2d898e	GOOD	0.0224	32,000
worst_bad	0.964	f863341e	GOOD	0.1647	32,000
worst_bad	0.965	9ad9967c	GOOD	0.0516	32,000
worst_bad	0.966	9fd07ca1	GOOD	0.0927	32,000
worst_bad	0.966	b6a84d2c	GOOD	0.046	32,000
worst_bad	0.966	d3a5a566	GOOD	0.0249	32,000
worst_bad	0.967	d5fb3a7f	GOOD	0.0816	32,000
worst_bad	0.968	4310eb0b	GOOD	0.0518	32,000
worst_bad	0.968	b89d22d0	GOOD	0.0429	32,000
worst_bad	0.969	7dcc112c	GOOD	0.0325	32,000
worst_bad	0.97	21e69136	GOOD	0.0911	32,000
worst_bad	0.974	94d1303a	GOOD	0.1462	32,000

emolia_filter_sr_32

60 аудиосэмплов для прослушивания и калибровки порогов фильтра качества.

Отсортированы от лучшего к худшему по bad_prob. SR: 24kHz.

Колонки

Колонка	Тип	Описание
`audio`	Audio	Аудиофайл (24kHz)
`category`	string	best_good / boundary_good / boundary_bad / worst_bad
`bad_prob`	float64	Вероятность плохого качества (0.0–1.0)
`sample_id`	string	ID сэмпла
`speaker_status`	string	GOOD / UNCERTAIN / BAD (speaker overlap)
`speaker_overlap`	float64	Максимальная вероятность overlap (0.0–1.0)
`sampling_rate`	int64	Оригинальный sample rate

Категории (по 15 файлов)

#	Категория	bad_prob	Описание
1–15	best_good	0.007–0.029	Лучшие сэмплы, далеко от порога
16–30	boundary_good	0.347–0.350	Good на грани порога (чуть ниже 0.35)
31–45	boundary_bad	0.350–0.353	Bad на грани порога (чуть выше 0.35)
46–60	worst_bad	0.963–0.974	Худшие сэмплы, далеко от порога

Фильтры

Quality filter — 3-канальный (LogisticRegression, 12 DSP-метрик):

Ch1: bad_prob >= 0.35 → bad
Ch2: spectral_autocorr_peak > 0.987 → bad (comb-filter)
Ch3: crest_factor_std < 1.20 → bad (transient)

Speaker overlap filter — Pyannote ONNX segmentation-3.0:

GOOD: max_overlap < 0.20
UNCERTAIN: 0.20–0.30
BAD: max_overlap >= 0.30

Downloads last month: 40

Number of rows:

Total file size:

35.9 MB