MARSA: Multi-accent Audio Recordings for Spanish Anti-spoofing — Augmented (×2) Release

MARSA (Multi-accent Audio Recordings for Spanish Anti-spoofing) is a corpus of 240,995 base utterances built from 35,927 bonafide recordings spanning seven Spanish accents across Spain and Latin America, 186,647 full-spoof utterances generated by six state-of-the-art open TTS voice-cloning systems, and 18,421 partial-spoof utterances in which one to three words of an otherwise bonafide recording are replaced by cloned speech using an energy-aware splicing engine. This record contains the augmented ×2 release tier: an acoustic-augmentation layer covering reverberation, additive noise, and telephony/VoIP codec degradation applied to the base corpus, released in ASVspoof2019 Logical Access protocol format (train/dev/eval splits, FLAC audio, protocol files, and per-sample metadata). This is one of several augmentation-factor releases planned for MARSA (×2 here; ×3 and ×5 to follow as separate records); the base (unaugmented) corpus and other factors are not included in this specific record. License note: MARSA's license is mixed at the per-sample level. Most of the corpus is released under CC BY-SA 4.0, but audio generated by the Fish-Speech (s1-mini) voice-cloning system is released under the more restrictive CC BY-NC-SA 4.0, following the terms of that system's own non-commercial model license. This record is labeled CC BY-NC-SA 4.0 at the record level to reflect the most restrictive component present. Each sample's generating system is recorded in its protocol/metadata entry, so users requiring an unrestricted commercial subset can mechanically exclude the Fish-Speech-derived portion. File reassembly: due to Zenodo's per-file size limit, the corpus archive is split into 15 sequential parts (MARSA_augmented_2x.zip.part000 through MARSA_augmented_2x.zip.part014). Download all 15 parts and concatenate them in numerical order into a single file named MARSA_augmented_2x.zip, then extract it with any standard zip tool. The reassembled archive is 30,419,258,700 bytes (~28.3 GiB).