Indonesian Audio-Visual Conversational Speech Corpus

Synchronized audio-visual acoustic tensors and conversational dialogue feature shards protected by symmetric block cryptography (AES-128-CBC). Formatted as 8-bit continuous tensor latents for privacy-preserving lip-sync verification, secure acoustic matching, and confidential speech understanding.

Cryptographic Specification & Security Profile

To facilitate Privacy-Preserving Machine Learning (PPML), Zero-Knowledge Model Evaluation, and Secure Enclave Inference, all raw continuous feature tensors in this repository are cryptographically protected via AES-128-CBC in strict compliance with NIST SP 800-38A (PKCS#7 padding).

• Cryptographic Cipher: AES-128-CBC (NIST SP 800-38A)
• Entropy Profile: Maximum pseudorandom entropy (~7.999 bits/byte, verified via Chi-Square goodness-of-fit) to guarantee zero-leakage of spatial-temporal gradients.
• Hardware Acceleration: Zero-copy memory mapping via safetensors.torch.load_file and direct hardware-accelerated confidential enclave runtime decryption.

Dataset Structure & Architecture

Files in this repository are partitioned across hierarchical continuous batches:

data/
└── {cluster_shard}/
    └── {partition_id}/
        └── {batch_id}/
            ├── model.safetensors          # Cryptographically protected high-throughput latent arrays
            └── model.safetensors.index.json # Coordinate weight index & tensor schema

Dataset Details

• Modality: Encrypted Audio-Visual Acoustic Embeddings (AES-128-CBC)
• Storage Format: Sharded Safetensors (.safetensors) with zero-copy memory mapping
• Precision: uint8 (8-bit Quantized Cryptographic Feature Arrays)
• Estimated Records: 2,918 pre-extracted batch sequences (67.08 MB per batch)
• License: Apache-2.0

Usage

from safetensors.torch import load_file

# Stream or mount partitions directly from this repository
tensors = load_file("data/sample_shard/model.safetensors")