Dolphin

Dolphin - Multilingual ASR Model

Dolphin is a cutting-edge multilingual, multitask Automatic Speech Recognition (ASR) model developed by DataoceanAI in collaboration with Tsinghua University. This model supports a wide variety of languages and dialects, enabling robust speech recognition across different linguistic backgrounds.

Key Features:

• Multilingual Support: Handles 40 Eastern languages and 22 Chinese dialects, making it suitable for diverse applications in different regions.
• Comprehensive Functionality: Capable of performing speech recognition, voice activity detection (VAD), segmentation, and language identification (LID).
• Innovative Architecture: Utilizes a joint CTC-Attention model architecture with Encoder and Decoder components for optimal ASR performance.
• Two-Level Language Token System: Enhances linguistic handling by incorporating region-specific tokens alongside language tokens.

Benefits:

• Diverse Language Processing: Ideal for applications needing support for various Eastern languages.
• High Accuracy: Trained on over 210,000 hours of data, ensuring high-quality output.
• Easy Integration: Simple installation and usage via Python, making it accessible for developers and researchers.

Dolphin represents a significant advancement in multilingual ASR, providing tools and solutions that cater to a global audience.