Datasets:

Modalities:
Audio
Libraries:
Datasets
License:
ChristophSchuhmann's picture
Public release with neutral metadata
1619701
Raw History Blame Contribute Delete
1.71 kB
{
"model": "Humaneness Podcast — Large 1.7 real-podcast full fine-tune",
"checkpoint": "step-00009874",
"step": 9874,
"tasks": 2400,
"counts": {
"reference_dialogue": 1200,
"invented_dialogue": 800,
"single_emotion": 400
},
"languages": {
"en": "English",
"de": "German",
"fr": "French",
"es": "Spanish",
"nl": "Dutch"
},
"cfg": 2.5,
"temperature": 1.0,
"top_p": 0.95,
"top_k": 50,
"repetition_penalty": 1.0,
"seeds": [
7101,
7102
],
"conditioning_cap_seconds": 60,
"durations_s": [
14,
20,
40,
56
],
"neutral_sentence_per_language": {
"en": "The package is on the table, and the meeting begins tomorrow morning.",
"de": "Das Paket liegt auf dem Tisch, und das Treffen beginnt morgen früh.",
"fr": "Le colis est sur la table et la réunion commence demain matin.",
"es": "El paquete está sobre la mesa y la reunión empieza mañana por la mañana.",
"nl": "Het pakket ligt op tafel en de vergadering begint morgenochtend."
},
"paired_general_ablation": "same literal text, speaker routing, references, duration, seed and CFG; GENERAL is blank or genuine/realistic/authentic",
"cfg_neutral_contract": "Preserve literal words, speaker routing, identity descriptions, durations and audio references/context. Neutralize acting/emotion directions and extra GENERAL delivery text. Audio codes are shared in both histories; conditional stop head is unchanged.",
"not_claimed": "Millisecond-accurate timing, independent verification of synthetic source gender, perfect 40-emotion realization, trained support for three-speaker podcast generation",
"created_unix": 1794041972.4490561
}