q1805/german-golden-audio_speech-IPA

🌟 German Golden Speech & IPA Corpus (FLEURS + Multilingual TEDx)

An ultra-clean, high-standard curated German speech dataset combining Google FLEURS (`de_de`) and Multilingual TEDx German (`mTEDx`), fully embedded with 16kHz WAV audio bytes, normalized orthographic text, and pre-computed International Phonetic Alphabet (IPA) transcriptions.

📊 Dataset Summary

• —Total Samples: 1,354 high-quality audio recordings.
• —Total Size: ~419 MB (Compressed Parquet format).
• —Audio Format: 16,000 Hz, Mono channel, 16-bit PCM.
• —Subsets & Origins:
• —`google_fleurs` (862 samples): Official test split from Google FLEURS. Studio-grade, noise-isolated read speech recorded by native German speakers across diverse topics.
• —`multilingual_tedx` (492 samples): Curated segments from German TEDx presentations. Captures natural pacing, spontaneous intonations, and authentic presentation acoustics with professionally verified human subtitles.

🚀 Key Use Cases

This dataset is designed as a multi-purpose golden asset for speech AI engineers and researchers:

1. 1.Rigorous Out-of-Domain Benchmarking:
2. 2.Evaluate ASR and Pronunciation Scoring models on both pristine studio acoustics (FLEURS) and natural live speeches (TEDx).
3. 3.Direct computation of Phoneme Error Rate (PER), Character Error Rate (CER), and Word Error Rate (WER).
4. 4.High-Quality Calibration & Fine-Tuning:
5. 5.Ideal for few-shot adaptation, domain adaptation, and acoustic calibration of models like HuBERT, Wav2Vec2, and Whisper.
6. 6.Pronunciation Assessment & GOP Scoring:
7. 7.Pre-aligned IPA target sequences make it immediately compatible with Goodness of Pronunciation (GOP) pipelines without external G2P dependencies.

💻 Quick Usage with 🤗 Datasets

1. Direct Streaming (Fast Evaluation)

from datasets import load_dataset

# Stream dataset instantly without downloading all files to disk
dataset = load_dataset("q1805/german-golden-benchmark", split="train", streaming=True)
sample = next(iter(dataset))

print("Source:", sample["source"])
print("Text:", sample["text"])
print("IPA:", sample["ipa"])
print("Audio array shape:", sample["audio"]["array"].shape)