humaneness-ears-base-medium / training_normalization.json
ChristophSchuhmann's picture
Release best Gemini-tuned Whisper Base and Small with code, normalization and evaluation
cd9b2d8 verified
Raw History Blame Contribute Delete
14.9 kB
{
"schema": "layered-multitask-training-normalization-v1",
"score_names": [
"emo_Affection",
"emo_Amusement",
"emo_Anger",
"emo_Astonishment_Surprise",
"emo_Awe",
"emo_Bitterness",
"emo_Concentration",
"emo_Confusion",
"emo_Contemplation",
"emo_Contempt",
"emo_Contentment",
"emo_Disappointment",
"emo_Disgust",
"emo_Distress",
"emo_Doubt",
"emo_Elation",
"emo_Embarrassment",
"emo_Emotional_Numbness",
"emo_Fatigue_Exhaustion",
"emo_Fear",
"emo_Helplessness",
"emo_Hope_Enthusiasm_Optimism",
"emo_Impatience_and_Irritability",
"emo_Infatuation",
"emo_Interest",
"emo_Intoxication_Altered_States_of_Consciousness",
"emo_Jealousy_and_Envy",
"emo_Longing",
"emo_Malevolence_Malice",
"emo_Pain",
"emo_Pleasure_Ecstasy",
"emo_Pride",
"emo_Relief",
"emo_Sadness",
"emo_Sexual_Lust",
"emo_Shame",
"emo_Sourness",
"emo_Teasing",
"emo_Thankfulness_Gratitude",
"emo_Triumph",
"vn_AGEV_reg",
"vn_AROU_reg",
"vn_ARSH_reg",
"vn_ATCK_reg",
"vn_BKGN_reg",
"vn_BRGT_reg",
"vn_CHNK_reg",
"vn_CLRT_reg",
"vn_COGL_reg",
"vn_DARC_reg",
"vn_DFLU_reg",
"vn_EMPH_reg",
"vn_ESTH_reg",
"vn_EXPL_reg",
"vn_FOCS_reg",
"vn_FULL_reg",
"vn_GEND_reg",
"vn_HARM_reg",
"vn_METL_reg",
"vn_RANG_reg",
"vn_RCQL_reg",
"vn_REGS_reg",
"vn_RESP_reg",
"vn_ROUG_reg",
"vn_R_CHST_reg",
"vn_R_HEAD_reg",
"vn_R_MASK_reg",
"vn_R_MIXD_reg",
"vn_R_NASL_reg",
"vn_R_ORAL_reg",
"vn_R_THRT_reg",
"vn_SMTH_reg",
"vn_STNC_reg",
"vn_STRU_reg",
"vn_S_ASMR_reg",
"vn_S_AUTH_reg",
"vn_S_CART_reg",
"vn_S_CASU_reg",
"vn_S_CONV_reg",
"vn_S_DRAM_reg",
"vn_S_FORM_reg",
"vn_S_MONO_reg",
"vn_S_NARR_reg",
"vn_S_NEWS_reg",
"vn_S_PLAY_reg",
"vn_S_RANT_reg",
"vn_S_STRY_reg",
"vn_S_TECH_reg",
"vn_S_WHIS_reg",
"vn_TEMP_reg",
"vn_TENS_reg",
"vn_VALN_reg",
"vn_VALS_reg",
"vn_VFLX_reg",
"vn_VOLT_reg",
"vn_VULN_reg",
"vn_WARM_reg",
"genuineness_0_6",
"blend_0_10",
"R_quality",
"eiv_extra:Age",
"eiv_extra:Arousal",
"eiv_extra:Authenticity",
"eiv_extra:Background_Noise",
"eiv_extra:Confident_vs._Hesitant",
"eiv_extra:Gender",
"eiv_extra:High-Pitched_vs._Low-Pitched",
"eiv_extra:Monotone_vs._Expressive",
"eiv_extra:Recording_Quality",
"eiv_extra:Serious_vs._Humorous",
"eiv_extra:Soft_vs._Harsh",
"eiv_extra:Submissive_vs._Dominant",
"eiv_extra:Valence",
"eiv_extra:Vulnerable_vs._Emotionally_Detached",
"eiv_extra:Warm_vs._Cold",
"eiv_extra:score_background_quality",
"eiv_extra:score_content_enjoyment",
"eiv_extra:score_overall_quality",
"eiv_extra:score_speech_quality",
"audiobox:CE",
"audiobox:CU",
"audiobox:PC",
"audiobox:PQ",
"dnsmos:SIG_raw",
"dnsmos:BAK_raw",
"dnsmos:OVRL_raw",
"dnsmos:SIG",
"dnsmos:BAK",
"dnsmos:OVRL",
"dnsmos:P808_MOS",
"burst_count_log1p",
"voiceclap_attribute:Affection",
"voiceclap_attribute:Age",
"voiceclap_attribute:Amusement",
"voiceclap_attribute:Anger",
"voiceclap_attribute:Arousal",
"voiceclap_attribute:Astonishment_Surprise",
"voiceclap_attribute:Authenticity",
"voiceclap_attribute:Awe",
"voiceclap_attribute:Background_Noise",
"voiceclap_attribute:Bitterness",
"voiceclap_attribute:Concentration",
"voiceclap_attribute:Confident_vs._Hesitant",
"voiceclap_attribute:Confusion",
"voiceclap_attribute:Contemplation",
"voiceclap_attribute:Contempt",
"voiceclap_attribute:Contentment",
"voiceclap_attribute:Disappointment",
"voiceclap_attribute:Disgust",
"voiceclap_attribute:Distress",
"voiceclap_attribute:Doubt",
"voiceclap_attribute:Elation",
"voiceclap_attribute:Embarrassment",
"voiceclap_attribute:Emotional_Numbness",
"voiceclap_attribute:Fatigue_Exhaustion",
"voiceclap_attribute:Fear",
"voiceclap_attribute:Gender",
"voiceclap_attribute:Helplessness",
"voiceclap_attribute:High-Pitched_vs._Low-Pitched",
"voiceclap_attribute:Hope_Enthusiasm_Optimism",
"voiceclap_attribute:Impatience_and_Irritability",
"voiceclap_attribute:Infatuation",
"voiceclap_attribute:Interest",
"voiceclap_attribute:Intoxication_Altered_States_of_Consciousness",
"voiceclap_attribute:Jealousy_&_Envy",
"voiceclap_attribute:Longing",
"voiceclap_attribute:Malevolence_Malice",
"voiceclap_attribute:Monotone_vs._Expressive",
"voiceclap_attribute:Pain",
"voiceclap_attribute:Pleasure_Ecstasy",
"voiceclap_attribute:Pride",
"voiceclap_attribute:Recording_Quality",
"voiceclap_attribute:Relief",
"voiceclap_attribute:Sadness",
"voiceclap_attribute:Serious_vs._Humorous",
"voiceclap_attribute:Sexual_Lust",
"voiceclap_attribute:Shame",
"voiceclap_attribute:Soft_vs._Harsh",
"voiceclap_attribute:Sourness",
"voiceclap_attribute:Submissive_vs._Dominant",
"voiceclap_attribute:Teasing",
"voiceclap_attribute:Thankfulness_Gratitude",
"voiceclap_attribute:Triumph",
"voiceclap_attribute:Valence",
"voiceclap_attribute:Vulnerable_vs._Emotionally_Detached",
"voiceclap_attribute:Warm_vs._Cold",
"voiceclap_attribute:duration",
"voiceclap_attribute:score_background_quality",
"voiceclap_attribute:score_content_enjoyment",
"voiceclap_attribute:score_overall_quality",
"voiceclap_attribute:score_speech_quality",
"voiceclap_attribute:talking_speed"
],
"score_mean": [
0.27749165892601013,
0.4777434766292572,
0.36061015725135803,
0.2777629792690277,
0.11817667633295059,
0.21946945786476135,
0.6705899834632874,
0.3177536427974701,
0.5056841969490051,
0.20877057313919067,
0.35425904393196106,
0.4897621273994446,
0.22247810661792755,
0.4589405953884125,
0.3137844204902649,
0.4519732594490051,
0.2580322027206421,
0.46182429790496826,
0.28273433446884155,
0.3217501640319824,
0.4309713840484619,
0.5629669427871704,
0.5035319924354553,
0.19843806326389313,
1.4420181512832642,
0.1724642664194107,
0.21464531123638153,
0.4350960850715637,
0.2800264358520508,
0.2569645941257477,
0.4043351113796234,
0.22033317387104034,
0.3699576258659363,
0.49465101957321167,
0.2548427879810333,
0.21115742623806,
0.21510320901870728,
0.34143486618995667,
0.318909615278244,
0.1931096464395523,
2.681518316268921,
2.4366743564605713,
2.5739870071411133,
1.9849865436553955,
2.7404820919036865,
1.9509913921356201,
1.8721773624420166,
1.894247055053711,
2.070889472961426,
1.7189459800720215,
2.307497978210449,
2.2513887882232666,
2.337125539779663,
0.11109915375709534,
2.223909378051758,
1.7921996116638184,
3.3836464881896973,
2.4652652740478516,
1.5194090604782104,
2.968099594116211,
2.737485408782959,
1.6384806632995605,
2.804781675338745,
2.0985488891601562,
2.074404716491699,
2.040480852127075,
1.722941517829895,
1.5071759223937988,
2.6270196437835693,
2.126116991043091,
1.9610368013381958,
1.8022409677505493,
2.57399582862854,
1.8595839738845825,
1.5977575778961182,
1.211488127708435,
1.7008559703826904,
3.7504734992980957,
2.533360481262207,
2.806027412414551,
1.269144892692566,
2.6660656929016113,
1.4076342582702637,
0.48746055364608765,
2.2685763835906982,
1.434207558631897,
2.3200457096099854,
0.528684139251709,
1.7669295072555542,
2.114962577819824,
1.6594414710998535,
2.150817632675171,
2.3777313232421875,
2.2164313793182373,
2.6844067573547363,
2.7245829105377197,
2.6052799224853516,
2.8253684043884277,
6.707726955413818,
1.1416735649108887,
3.199366569519043,
1.381827712059021,
2.681858539581299,
0.7440048456192017,
1.0924224853515625,
-0.11666019260883331,
1.838754415512085,
1.840975046157837,
2.1304681301116943,
0.9606292843818665,
0.3583308458328247,
0.3837394118309021,
0.07115401327610016,
1.7351616621017456,
0.11160333454608917,
3.062971591949463,
4.46293830871582,
2.5364186763763428,
1.7876968383789062,
4.791499614715576,
5.558483123779297,
3.0696988105773926,
6.888725757598877,
3.7026989459991455,
3.1376471519470215,
2.988908529281616,
3.332533121109009,
3.183006763458252,
2.728473424911499,
3.371939182281494,
0.564864993095398,
1.0070288181304932,
3.0217714309692383,
0.5854920148849487,
1.0566563606262207,
1.4644229412078857,
0.6210613250732422,
2.788144588470459,
0.5391162037849426,
0.5500497817993164,
0.6080928444862366,
1.1852548122406006,
1.1811816692352295,
0.6684321761131287,
0.8097957968711853,
0.7250347137451172,
0.6182515025138855,
0.7696337699890137,
0.6267121434211731,
0.9318065047264099,
0.5428418517112732,
0.6230013370513916,
0.37965333461761475,
0.6574907302856445,
0.4837093651294708,
0.6278012990951538,
-0.01745273359119892,
0.6736047267913818,
1.8283263444900513,
0.7801159620285034,
1.1286685466766357,
0.6222699284553528,
1.5985472202301025,
0.5895588994026184,
0.531328558921814,
0.5524753332138062,
0.8625041842460632,
1.8350659608840942,
0.3901818096637726,
0.5491418838500977,
0.7273247241973877,
2.2944633960723877,
1.0110139846801758,
0.7869164943695068,
0.972858726978302,
0.6760010123252869,
0.5549875497817993,
0.18813416361808777,
0.7186443209648132,
0.6201862096786499,
0.614105761051178,
1.3171813488006592,
0.7256725430488586,
-0.03641652688384056,
1.678699254989624,
-0.03192274272441864,
11.635364532470703,
3.2266478538513184,
4.694666862487793,
2.603773593902588,
1.9127005338668823,
9.481627464294434
],
"score_std": [
0.5437366962432861,
0.8676553964614868,
0.5801360011100769,
0.5521981716156006,
0.4197714924812317,
0.4069267213344574,
0.7305826544761658,
0.5334029793739319,
0.5889452695846558,
0.45113658905029297,
0.5038090348243713,
0.7013412714004517,
0.49026989936828613,
0.8173884749412537,
0.48334044218063354,
0.7184596061706543,
0.4370509088039398,
0.9614661931991577,
0.5407941937446594,
0.611741304397583,
0.7080591320991516,
0.7584722638130188,
0.6841333508491516,
0.5412347912788391,
0.9140985608100891,
0.5562154054641724,
0.44575998187065125,
0.6358309984207153,
0.5236847996711731,
0.5751715898513794,
0.6105303764343262,
0.51991206407547,
0.6979016065597534,
0.7643030285835266,
0.4977344572544098,
0.5240214467048645,
0.43236520886421204,
0.5380140542984009,
0.6912878751754761,
0.5280161499977112,
1.2599846124649048,
1.126529574394226,
0.8762896060943604,
1.3167362213134766,
0.7566960453987122,
0.8106063008308411,
1.2465068101882935,
0.9609140753746033,
1.0542023181915283,
1.362148642539978,
1.1811630725860596,
1.571174144744873,
0.7787088751792908,
0.3274502456188202,
1.109856367111206,
0.883672297000885,
1.5924972295761108,
0.8279261589050293,
0.9576501846313477,
0.891509473323822,
0.9002096056938171,
1.2561882734298706,
1.0695379972457886,
0.8817092776298523,
0.8490350842475891,
0.8235664367675781,
1.023396611213684,
0.8368624448776245,
1.0560996532440186,
0.7440944910049438,
0.8052754402160645,
1.111371636390686,
1.13865327835083,
0.9860734343528748,
1.3787169456481934,
1.0388771295547485,
1.3691930770874023,
1.2758985757827759,
1.305350422859192,
1.1393134593963623,
1.008335828781128,
1.2723127603530884,
1.4352364540100098,
0.5510048270225525,
1.2999988794326782,
1.275938868522644,
1.39365553855896,
1.217934250831604,
1.5986617803573608,
1.2006604671478271,
0.8322792053222656,
1.1298775672912598,
1.0982979536056519,
1.2067712545394897,
0.8684567809104919,
0.9905231595039368,
0.9048547744750977,
1.6066662073135376,
2.717972755432129,
0.3311099708080292,
0.2950047552585602,
0.4359813630580902,
0.25,
0.26394614577293396,
0.3041084110736847,
0.9973793625831604,
0.25,
0.47023749351501465,
0.33257871866226196,
0.34091120958328247,
0.3194079101085663,
0.32314765453338623,
0.38162848353385925,
0.43580567836761475,
0.29992207884788513,
0.6935931444168091,
0.30376729369163513,
0.5075397491455078,
0.25,
1.0166618824005127,
0.9988844394683838,
1.6154870986938477,
0.8792684674263,
0.6690715551376343,
1.1252458095550537,
0.8129898905754089,
0.5011401772499084,
0.9285669922828674,
0.6066545248031616,
0.4964241683483124,
0.6485206484794617,
0.5545063614845276,
0.5437508821487427,
0.5738969445228577,
0.5073710680007935,
0.6225984692573547,
0.4788701832294464,
0.25,
0.25941038131713867,
0.2641003727912903,
0.25,
0.5305075645446777,
0.46878623962402344,
0.3352125883102417,
0.4053431451320648,
0.33969345688819885,
0.30695465207099915,
0.29053059220314026,
0.25,
0.3499687612056732,
0.25,
0.39873725175857544,
0.31471413373947144,
0.26617875695228577,
0.2603147029876709,
0.25,
1.0046077966690063,
0.2682698369026184,
0.3735311031341553,
0.3972975015640259,
0.542029619216919,
0.3326053023338318,
0.43933433294296265,
0.4639695882797241,
0.25,
0.25,
0.3459997773170471,
0.5964633822441101,
0.25,
0.3012017607688904,
0.34898921847343445,
0.3281205892562866,
0.4159489870071411,
0.3129201829433441,
0.5704670548439026,
0.4178532063961029,
0.2817857563495636,
0.3601606488227844,
0.29048749804496765,
0.35900384187698364,
0.3793472647666931,
0.5293529033660889,
0.3587714433670044,
0.6340725421905518,
0.5558832883834839,
0.3993605971336365,
6.333652019500732,
0.49855202436447144,
0.25,
0.41230252385139465,
0.25,
3.752267599105835
],
"cps_mean": 18.207849092730562,
"cps_std": 7.742389961392013,
"score_formula": "raw = network_output * score_std + score_mean",
"cps_formula": "characters_per_second = network_output * cps_std + cps_mean"
}