Humaneness-Voice-Large-1.3-Alpha-Listening
收藏资源简介:
# Humaneness Voice Large1.3Alpha: matched listening and fictional emotion masking 270new final-SFT-step9,799predictions:70matched predictions on the exact previous1.2demonstration cases, plus200new masked-acting/control predictions. This is NOT an RL model or an RL evaluation. The next33%CE/33%top2-emotion/ 34%genuineness RL recipe is a draft awaiting listening review and a fresh smoke. Every new condition has one fixed seed, not a best-of-N or duration-selected take. Actual durations are retained, with no trimming or time-stretching. Matching1.2/1.3inputs include literal text, directing prompt, requested frame budget, actual baseline seed and identical separate12-second reference codes. Masking is fictional acting: a concealed feeling versus an outward performance. The separate cue-removal control removes explicit emotional instructions, not words or timing. Maintained masks and brief mask breaks are labelled explicitly. Do not interpret fictional subtext as measured private mental states. audio/contains individually streamable128kbps mono MP3s. predictions.tar packs the same MP3s, all12MOSS-v2 indices per80ms frame, and full per-output JSON. predictions.json/TASKS.json preserve exact prompts, seeds and model provenance. The task/source contract and local export proof are retained with the run. The unchanged1.2audio and the five public-synthetic Gemini reference recordings remain in laion/Humaneness-Voice-Large-1.2-Alpha-Demo-Audio. references.json and references.npz retain their provenance and exact codec arrays. No human reference audio is republished by this study. Newly written scenes, predictions and code: CC BY4.0, LAION contributors,2026. Source reference attribution remains with laion/german-english-voice-acting-challenges. Respect consent and upstream terms. No CE, emotion, genuineness or speaker-similarity improvement is established by this listening page alone. Current summary values cover durations and runtime. Listening Space: https://laion-humaneness-voice-large-1-2-alpha-listening.static.hf.space/



