Tuning parameters
56B
tokens in model
0.7M
train-hours of speech
57
languages to listen
16
languages to speak
Chatbot who listens
Recognize
The way you ask all exciting questions
1.
56 billion tokens and 700,000 training hours AI model resulting in cutting-edge speech recognition
2.
Contextual understanding that goes beyond simple keywords, allowing for more natural, engaging conversations, and personalized interaction
3.
Professional fluency in 57 diverse languages with a wide range from English to Mandarin Chinese
Chatbot who speaks
1.
More than 20 unique multilingual AI voices trained with help of experienced voice actors to sound real
2.
Adaptive tone speaking style that adjusts to context of conversation, mirroring your emotions and energy levels, keeping conversation flowing naturally
3.
Customize scenario script of the model persona's backstory to get more immersive conversations
Answer
In a way you can easily listen & understand
Project roadmap
From
11
March
2025
To
9
June
2025
Phase I: Multilingual foundation
Core for speech recognition & NLP engine
1.
Diverse language dataset of speech and text in 57 languages with representation of various accents, dialects, and speaking styles
2.
Transformer-based architecture for speech recognition and NLP tasks, capable of handling multilingual input & large-scale training
3.
Evaluation metrics, hyperparameter tuning, and data augmentation to achieve professional fluency & accuracy across all 57 languages
From
10
June
2025
To
8
September
2025
Phase II: Personality customization
Emotions, adaptive tone & speaking style
1.
Sentiment analysis capabilities integration allows model to detect and interpret emotions expressed in text or speech accurately
2.
A wide library of unique AI voices prepared by experienced voice actors representing diverse language backgrounds & acting styles
3.
Tool for defining and customizing AI personalities allows users to select or create personalities that align with their preferences
From
9
September
2025
To
8
December
2025
Phase III: Contextual awareness
Topics perception & proactive suggestions
1.
Coreference resolution capabilities integration to accurately track entities across multiple sentences, ensuring reference accuracy
2.
Algorithm to track evolution of topics throughout the conversation, recognizing shifts in focus and adapting its responses accordingly
3.
System to retrieve and present personalized content like articles, recommendations, and data based on user's interests and context
From
9
December
2025
To
9
March
2026
Phase IV: Advanced interaction
Seamless language, text format switching
1.
Cross-lingual transfer learning to enhance the AI model's ability to understand and respond appropriately in multilingual conversations
2.
Normalization of text input to handle variations in spelling, punctuation, and grammar and allowing for consistent processing
3.
Switching engine for seamless transitions between languages and text formats in real-time, without interrupting flow of conversation