Datasetpublic

playwithmino/aishell1mix-ver2-n100-per-mix

AISHELL-1 Mix ver2 — 100 clips per mix This is AISHELL-1 Mix ver2, not ver1. 8 kHz mono test subset: 100 mixtures per speaker count (N=1\ldots5) (50 mix_clean + 50 mix_both each) → 500 clips. Derived from the local aishell1mix_ver2 test SCPs (data/scp/scp_aishell1mix_ver2). Includes mixture + oracle speaker stems and transcripts. Split Count 1mix / 2mix / 3mix / 4mix / 5mix 100 each clean / both 250 each Files manifests/test.jsonl —… See the full description on the dataset page: https://huggingface.co/datasets/playwithmino/aishell1mix-ver2-n100-per-mix.

sourceHugging Faceotherupdated 26d agoView on Hugging Face

0likes71downloads

Dataset Card

AISHELL-1 Mix ver2 — 100 clips per mix

This is AISHELL-1 Mix ver2, not ver1. 8 kHz mono test subset: 100 mixtures per speaker count \(N=1\ldots5\) (50 mix_clean + 50 mix_both each) → 500 clips.

Derived from the local aishell1mix_ver2 test SCPs (data/scp/scp_aishell1mix_ver2). Includes mixture + oracle speaker stems and transcripts.

Split	Count
1mix / 2mix / 3mix / 4mix / 5mix	100 each
clean / both	250 each

Files

—manifests/test.jsonl — scorer-compatible rows; mix_path / sources[].audio_path are relative to the repo root (wav8k/...)
—wav8k/clean/{mix,s1..s5}/ and wav8k/both/{mix,s1..s5}/

Load

bash

huggingface-cli download playwithmino/aishell1mix-ver2-n100-per-mix \
  --repo-type dataset --local-dir data/aishell1mix_ver2_n100_per_mix

This subset is for research evaluation. Original AISHELL-1 / AISHELL-1 Mix licenses still apply.