Datasets:
license: mit
dataset_info:
features:
- name: utt_id
dtype: string
- name: path
dtype: audio
- name: label
dtype: string
- name: source
dtype: string
- name: source_text
dtype: string
- name: source_speaker_id
dtype: string
- name: replay_details
struct:
- name: room_size
dtype: string
- name: player
dtype: string
- name: recorder
dtype: string
- name: distance
dtype: string
- name: synthesis_details
struct:
- name: model
dtype: string
- name: reference
dtype: string
- name: reference_text
dtype: string
- name: reference_speaker_id
dtype: string
splits:
- name: train
num_bytes: 1881368369.834
num_examples: 39926
- name: dev
num_bytes: 190120550.729
num_examples: 3973
- name: closed_set_eval
num_bytes: 276895281.202
num_examples: 5991
- name: open_set_eval
num_bytes: 1199943251
num_examples: 25600
download_size: 3263822188
dataset_size: 3548327452.7650003
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
- split: dev
path: data/dev-*
- split: closed_set_eval
path: data/closed_set_eval-*
- split: open_set_eval
path: data/open_set_eval-*
EchoFake: A Replay-Aware Dataset for Practical Speech Deepfake Detection
Paper link: http://arxiv.org/abs/2510.19414
Code for baseline models is available at https://github.com/EchoFake/EchoFake
Auto-recording tools is available at https://github.com/EchoFake/EchoFake/tree/main/tools
Abstract
The growing prevalence of speech deepfakes has raised serious concerns, particularly in real-world scenarios such as telephone fraud and identity theft. While many anti-spoofing systems have demonstrated promising performance on laboratory-generated synthetic speech, they often fail when confronted with physical replay attacks—a common and low-cost form of attack used in practical settings. Our experiments show that models trained on existing datasets exhibit severe performance degradation, with average accuracy dropping to 59.6% when evaluated on replayed audio. To bridge this gap, we present EchoFake, a comprehensive dataset comprising more than 120 hours of audio from over 13,000 speakers, featuring both cutting-edge zero-shot text-to-speech (TTS) speech and physical replay recordings collected under varied device configurations and real-world environmental settings. Additionally, we evaluate three baseline detection models and show that models trained on EchoFake achieve lower average EERs across datasets, indicating better generalization. By introducing more practical challenges relevant to real-world deployment, EchoFake offers a more realistic foundation for advancing spoofing detection methods.