---
title: Humaneness-Voice-Large-1.3-Alpha-Listening
canonical_url: "https://www.modelscope.cn/datasets/laion/Humaneness-Voice-Large-1.3-Alpha-Listening"
md_url: "https://www.modelscope.cn/datasets/laion/Humaneness-Voice-Large-1.3-Alpha-Listening.md"
repository: laion/Humaneness-Voice-Large-1.3-Alpha-Listening
last_updated: 2026-10-24
license: cc-by-4.0
storage_size: "198 MB"
downloads: 90
stars: 0
---

# Humaneness-Voice-Large-1.3-Alpha-Listening

> Humaneness-Voice-Large-1.3-Alpha-Listening - laion 在 ModelScope 开源的数据集。Humaneness Voice Large1.3Alpha: matched listening and fictional emotion masking

laion/Humaneness-Voice-Large-1.3-Alpha-Listening 是 ModelScope 魔搭社区上的数据集，存储大小 198 MB，采用 cc-by-4.0 许可。

- **Repository**: laion/Humaneness-Voice-Large-1.3-Alpha-Listening
- **License**: cc-by-4.0
- **Storage size**: 198 MB
- **Downloads**: 90
- **Stars**: 0
- **Last updated**: 2026-10-24

Source: https://www.modelscope.cn/datasets/laion/Humaneness-Voice-Large-1.3-Alpha-Listening

---

# 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/