Datasetpublic

Aynursusuz/tts-zh-zonos2-expressive

ZONOS2 — Accurate vs Expressive (Mandarin voice cloning) Side-by-side A/B comparison of Zyphra/ZONOS2 accurate mode (accurate_mode=true) vs expressive mode (accurate_mode=false). Same reference voice and same target text per row, cloned twice — one per mode — so each can be heard back to back. Reference voices are clean Qwen3 generations. Columns column meaning index row id ref_text text of the reference voice ref_audio reference voice (cloning… See the full description on the dataset page: https://huggingface.co/datasets/Aynursusuz/tts-zh-zonos2-expressive.

sourceHugging Faceupdated 3mo agoView on Hugging Face

0likes18downloads

Dataset Card

ZONOS2 — Accurate vs Expressive (Mandarin voice cloning)

Side-by-side A/B comparison of Zyphra/ZONOS2 accurate mode (accurate_mode=true) vs expressive mode (accurate_mode=false).

Same reference voice and same target text per row, cloned twice — one per mode — so each can be heard back to back. Reference voices are clean Qwen3 generations.

Columns

column	meaning
`index`	row id
`ref_text`	text of the reference voice
`ref_audio`	reference voice (cloning source)
`clone_text`	target text spoken in both clones
`clone_accurate`	clone, accurate mode
`clone_expressive`	clone, expressive mode
`dnsmos_accurate`	DNSMOS P.835 OVRL of the accurate clone
`dnsmos_expressive`	DNSMOS P.835 OVRL of the expressive clone
`keep_accurate`	`True` if `dnsmos_accurate >= 3.0` (quality filter)
`keep_expressive`	`True` if `dnsmos_expressive >= 3.0` (quality filter)

DNSMOS measures signal quality only — not expressiveness/prosody, which is best judged by ear.