Datasets:
id stringlengths 22 22 | speaker_embedding listlengths 2.05k 2.05k |
|---|---|
EN_5fwBwAdPk5A_W000049 | [
-0.198486328125,
0.029754638671875,
-0.296142578125,
0.2012939453125,
0.039764404296875,
0.1470947265625,
-0.349853515625,
-5.171875,
0.39111328125,
-0.331787109375,
-0.07928466796875,
-0.10040283203125,
-0.06146240234375,
-0.036651611328125,
-0.031036376953125,
0.2091064453125,
-0.0... |
EN_5fwBwAdPk5A_W000092 | [
-0.12152099609375,
-0.040191650390625,
-0.34228515625,
0.06488037109375,
0.09417724609375,
0.25146484375,
-0.26953125,
-4.89453125,
0.413330078125,
-0.343994140625,
-0.0011034011840820312,
-0.06939697265625,
-0.0207061767578125,
0.013336181640625,
-0.049530029296875,
0.231689453125,
... |
EN_5fwBwAdPk5A_W000012 | [
-0.1839599609375,
0.007537841796875,
-0.40087890625,
0.12841796875,
0.0233154296875,
0.197509765625,
-0.275146484375,
-5.109375,
0.392333984375,
-0.482666015625,
-0.07745361328125,
-0.08935546875,
-0.01280975341796875,
-0.093505859375,
0.01171875,
0.25439453125,
-0.1307373046875,
-... |
EN_5fwBwAdPk5A_W000104 | [
-0.1898193359375,
-0.05859375,
-0.39794921875,
0.1568603515625,
-0.0089569091796875,
0.1378173828125,
-0.301513671875,
-5.0078125,
0.41650390625,
-0.43603515625,
-0.03411865234375,
-0.0677490234375,
-0.08453369140625,
-0.030120849609375,
-0.0171051025390625,
0.215087890625,
-0.114685... |
EN_5fwBwAdPk5A_W000076 | [
-0.161865234375,
0.001033782958984375,
-0.372314453125,
0.10205078125,
0.0234527587890625,
0.1627197265625,
-0.3271484375,
-5.078125,
0.38525390625,
-0.39208984375,
-0.053985595703125,
-0.08935546875,
-0.036712646484375,
-0.034637451171875,
0.01372528076171875,
0.2283935546875,
-0.15... |
EN_5fwBwAdPk5A_W000113 | [-0.1343994140625,-0.004058837890625,-0.4150390625,0.075927734375,0.06329345703125,0.25537109375,-0.(...TRUNCATED) |
EN_5fwBwAdPk5A_W000138 | [-0.17822265625,0.01036834716796875,-0.382568359375,0.12353515625,0.038330078125,0.1865234375,-0.314(...TRUNCATED) |
EN_5fwBwAdPk5A_W000150 | [-0.1578369140625,-0.01104736328125,-0.38818359375,0.079345703125,0.044830322265625,0.2034912109375,(...TRUNCATED) |
EN_5fwBwAdPk5A_W000159 | [-0.1444091796875,0.01519012451171875,-0.41943359375,0.07989501953125,0.009185791015625,0.2342529296(...TRUNCATED) |
EN_5fwBwAdPk5A_W000174 | [-0.1029052734375,-0.05169677734375,-0.400390625,0.08148193359375,0.058441162109375,0.210205078125,-(...TRUNCATED) |
End of preview. Expand in Data Studio
Emilia-YODAS English Qwen3-TTS Speaker Embeddings
This dataset contains precomputed speaker embeddings for the English subset of Emilia-YODAS. Each row maps an Emilia-YODAS sample ID to one speaker embedding extracted from the corresponding audio.
Dataset Details
- Source dataset:
amphion/Emilia-Dataset - Source subset: Emilia-YODAS English
- Embedding model:
Qwen/Qwen3-TTS-12Hz-1.7B-Base - Embedding shape:
(2048,) - Embedding dtype:
float16 - Rows: 4,516,833
- Split:
train - Additional local-access format:
arrow/ - Columns:
id: source Emilia-YODAS sample IDspeaker_embedding: 2048-dimensional speaker embedding
Loading
from datasets import load_dataset
dataset = load_dataset(
"duplexio/emilia-yodas-en-speaker-embeddings",
split="train",
)
For local memory-mapped access, download the repository snapshot and load the Arrow copy:
from datasets import load_from_disk
from huggingface_hub import snapshot_download
root = snapshot_download("duplexio/emilia-yodas-en-speaker-embeddings")
dataset = load_from_disk(f"{root}/arrow")
The arrow/ directory also includes a sorted ID index:
import numpy as np
sorted_ids = np.load(f"{root}/arrow/id_index.sorted_ids.npy", mmap_mode="r")
row_indices = np.load(f"{root}/arrow/id_index.row_indices.npy", mmap_mode="r")
sample_id = "EN_5fwBwAdPk5A_W000049"
key = sample_id.encode("utf-8")
pos = np.searchsorted(sorted_ids, key)
row = int(row_indices[pos])
embedding = dataset[row]["speaker_embedding"]
Intended Use
These embeddings are intended for speech model pretraining, speaker conditioning, speaker retrieval, or analysis workflows that need per-utterance speaker representations for Emilia-YODAS English audio.
Limitations
The dataset contains embeddings only, not the original audio or transcripts. Embedding quality depends on the upstream audio quality and on the embedding model. These embeddings should not be treated as verified speaker identities.
- Downloads last month
- 185