GigaSpeech 3:以弱標註數據擴展多方言阿拉伯語語音識別
GigaSpeech 3: Scaling Multi-Dialect Arabic Speech Recognition with Weakly Labeled Data
GigaSpeech 3 建立涵蓋阿爾及利亞、埃及、伊拉克、沙特及敍利亞阿拉伯語的弱標註語音語料庫,包含 39,000 小時及 10 million 段語音。
Published on Nov 3
Authors:
,
,
,
,
,
,
,
,
,
,
,
,
,
,
Abstract
GigaSpeech 3 provides 39,000 hours of weakly transcribed dialectal Arabic speech, enabling improved recognition across five dialects via self-supervised labeling and noisy-student refinement.
Dialectal Arabic speech recognition remains constrained by the scarcity of dialect-resolved transcribed speech despite the abundance of web audio. We introduce GigaSpeech 3, a corpus of 39,000 hours and 10 million weakly transcribed speech segments covering Algerian, Egyptian, Iraqi, Saudi, and Syrian Arabic. Starting from 60 hours of manual transcription per dialect, we train labeling models using speech-recognition-biased self-supervised learning and apply them to dialect-matched web audio. Data selection and four rounds of noisy-student refinement for Saudi and Syrian Arabic yield the final corpus. We validate its utility through dialect-specific and multi-dialect speech recognition training and cross-domain evaluation. On GigaSpeechBench, the final dialect-specific models outperform their labeling models across all five dialects and achieve a macro-average word error rate of 23.86%, a 16.6% relative reduction from the strongest third-party system covering all five dialects, while outperforming the best third-party result on three dialects. A multi-dialect model using self-predicted dialect identities achieves a macro-average word error rate of 25.51% while also supporting Modern Standard Arabic. Across eight conditions from three external corpora, a GigaSpeech 3-trained Zipformer achieves the lowest word error rate among six evaluated models in every condition, while adapting the Qwen3 speech recognizer on GigaSpeech 3 improves performance across all eight conditions. We release the corpus and recipes as a foundation for dialect-aware Arabic speech processing.
View arXiv page View PDF Project page GitHub Add to collection
🎙️ GigaSpeech 3
Scaling Multi-Dialect Arabic Speech Recognition with Weakly Labeled Data
GigaSpeech 3 is a large-scale, weakly labeled Arabic speech corpus containing approximately 10 million speech segments across Algerian (DZA), Egyptian (EGY), Iraqi (IRQ), Saudi (SAU), and Syrian (SYR) Arabic. It supports dialect-specific and multi-dialect automatic speech recognition (ASR), weak supervision, and low-resource speech research.
The accompanying manuscript reports approximately 39,000 hours for its study corpus. Starting from only 51–63 hours of manual transcription per dialect, GigaSpeech 3 combines ASR-biased self-supervised learning, automatic transcription, data selection, and noisy-student refinement to build reusable audio–text supervision.
📥 Download and Demo
- 📄 Paper: arXiv · PDF
- 🤗 Training data: GigaSpeech 3 on Hugging Face
- 🎧 Online demo: Try GigaSpeech 3 Arabic ASR
- 🧪 Evaluation data: GigaSpeechBench on Hugging Face
- 🏆 Evaluation toolkit and leaderboard: SpeechColab/GigaSpeechBench
See the dataset card for file descriptions and terms of use. The current GitHub repository provides the project overview; training recipes and model checkpoint links will be added as they become available.
🌍 Dataset
🗣️ Dialects
| Code | Dialect |
|---|---|
| DZA | Algerian Arabic |
| EGY | Egyptian Arabic |
| IRQ | Iraqi Arabic |
| SAU | Saudi Arabic |
| SYR | Syrian Arabic |
Each dialect is released as a weakly labeled training partition. The manual seed data and MSA data used in the manuscript are separate from these five partitions. Evaluation uses independently transcribed data from GigaSpeechBench and external Arabic corpora.
Get this paper in your agent:
hf papers read 2611.02476
Don't have the latest CLI?
curl -LsSf https://hf.co/cli/install.sh | bash
Models citing this paper 0
No model linking this paper
Cite arxiv.org/abs/2611.02476 in a model README.md to link it from this page.
Datasets citing this paper 0
No dataset linking this paper
Cite arxiv.org/abs/2611.02476 in a dataset README.md to link it from this page.
Spaces citing this paper 0
No Space linking this paper
Cite arxiv.org/abs/2611.02476 in a Space README.md to link it from this page.
Collections including this paper 0
No Collection including this paper
Add this paper to a collection to link it from this page.
來源:HuggingFace Daily Papers(社區熱門論文) · huggingface.co