You need to agree to share your Hugging Face information to access this model

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Log in or Sign Up to review the conditions and access this model content.

๐Ÿ›๏ธ Axom AI Master - XUWOR (เฆธเงเฆเงฑเงฐ)

Assam's First Sovereign Full-Duplex Multimodal Artificial Intelligence Ecosystem


๐ŸŒŸ The Historic Milestone for Assam

Until today, developing state-of-the-art AI for Assam meant stitching together disjointed, unreliable third-party APIs that failed to comprehend the cultural spirit, phonetic subtleties, and linguistic richness of Assamese.

Axom AI Master - XUWOR (เฆธเงเฆเงฑเงฐ) changes the paradigm forever.

It is Assamโ€™s first complete sovereign conversational AI ecosystem. Xuwor acts as the master conductor, harmonizing four proprietary foundation models into a unified full-duplex conversational experience:

  1. ๐ŸŽ™๏ธ Axom AI Xuna (เฆถเงเฆจเฆพ): 600M Conformer ASR Engine (Hearing & Transcription)
  2. ๐Ÿง  Axom AI Progya (เฆชเงเงฐเฆœเงเฆžเฆพ): 2.00B Dense Transformer Brain (Reasoning & Comprehension)
  3. ๐Ÿ”Š Axom AI Puhor (เฆชเง‹เฆนเงฐ): 350M Diffusion Flow-Matching Voice (Warm Human Speech)
  4. ๐ŸŒ‰ Axom AI Xetu (เฆธเง‡เฆคเง): 200M Neural Translation Bridge (English โ†” Assamese Connection)

๐Ÿ”„ Unified Ecosystem Architecture

                      โ”Œโ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”
                      โ”‚ Spoken Assamese Input  โ”‚
                      โ””โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”ฌโ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”˜
                                  โ”‚
                                  โ–ผ
            โ”Œโ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”
            โ”‚  1. AXOM AI โ€” XUNA (เฆถเงเฆจเฆพ)                    โ”‚
            โ”‚  600M Hybrid Conformer-CTC ASR Engine         โ”‚
            โ”‚  โ€ข 1.29s latency on standard CPU (RTF 0.261x) โ”‚
            โ”‚  โ€ข Monsoon rain & telephony noise resilient   โ”‚
            โ””โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”ฌโ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”˜
                                  โ”‚ Recognized Native Text
                                  โ–ผ
            โ”Œโ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”
            โ”‚  2. AXOM AI โ€” PROGYA (เฆชเงเงฐเฆœเงเฆžเฆพ)                  โ”‚
            โ”‚  2.00 Billion Parameter Dense LLM Brain       โ”‚
            โ”‚  โ€ข 68,096 Assamese vocabulary tokenization    โ”‚
            โ”‚  โ€ข Grounded in 6.19M Assamese cultural recordsโ”‚
            โ””โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”ฌโ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”˜
                                  โ”‚ Intelligent Response Text
                                  โ–ผ
            โ”Œโ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”
            โ”‚  3. AXOM AI โ€” PUHOR (เฆชเง‹เฆนเงฐ)                    โ”‚
            โ”‚  350M Zero-Shot Flow-Matching Voice Engine    โ”‚
            โ”‚  โ€ข Studio 24,000 Hz human-grade speech        โ”‚
            โ”‚  โ€ข Calibrated conversational cadence          โ”‚
            โ””โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”ฌโ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”˜
                                  โ”‚
                                  โ–ผ
                      โ”Œโ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”
                      โ”‚ Spoken Assamese Output โ”‚
                      โ””โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”˜

๐ŸฅŠ The Paradigm Shift: Why Xuwor Dominates

Feature / Dimension Fragmented Third-Party Stacks ๐Ÿ›๏ธ Axom AI โ€” Xuwor
Multimodal Integration Disconnected APIs with compounding errors ๐Ÿ”„ End-to-End unified full-duplex orchestration
Speech Latency 8.0s - 15.0s cloud roundtrips โšก Real-time edge execution on standard 8 GB CPU
Phonetic & Cultural Grounding โŒ Generic Western/Hindi/Bengali bias ๐ŸŒพ 100% Khฤแนญi Asamiya phonology & Buranji wisdom
Data Privacy & Governance Data exposed to external corporate clouds ๐Ÿ›ก๏ธ Strictly private, sovereign, air-gapped deployment
Memory Optimization 24 GB+ GPU VRAM clusters required ๐Ÿ’Ž Under 3.8 GB peak RAM for the compound system

๐Ÿ“ฆ Connected Private Foundation Repositories

Engine Official Repository Parameter Size Core Capability
Voice Synthesis (TTS) axom-ai/axomai-voice-puhor 350 Million Studio 24kHz Flow-Matching DiT
Speech Recognition (ASR) axom-ai/axomai-speech-xuna 600 Million 0.261x RTF Conformer-CTC ONNX
Cognitive Brain (LLM) axom-ai/axomai-brain-progya 2.00 Billion 68k Vocab Dense Assamese Foundation
Translation Bridge axom-ai/axomai-trans-xetu 200 Million Bidirectional English โ†” Assamese

๐Ÿš€ Quick Start: Unified Master Python API

from pipeline import AxomAI

# 1. Initialize the Complete Ecosystem
ai = AxomAI(token="your_hf_access_token")

# 2. Run Full-Duplex Conversational Speech-to-Speech
conversation = ai.chat(audio_input_path="user_voice.wav", output_path="ai_reply.wav")
print("Heard (Xuna):", conversation["user_transcript"])
print("Thought (Progya):", conversation["ai_response_text"])
print("Spoke (Puhor):", conversation["output_audio_path"])

# 3. Individual Foundation Engine Invocations
transcript  = ai.listen("recording.wav")           # Speech-to-Text (Xuna)
narrative   = ai.think("เฆเฆฆเฆฟเฆจ เฆ…เฆธเฆฎเงฐ เฆเฆ–เฆจ เฆ—เฆพเฆเงฑเฆค")      # Cognitive Brain (Progya)
audio_file  = ai.speak("เฆจเฆฎเฆธเงเฆ•เฆพเงฐ, เฆฎเฆ‡ เฆ…เฆธเฆฎ เฆเฆ†เฆ‡เฅค")     # Voice Synthesis (Puhor)
translation = ai.translate("Hello Assam")         # Translation Bridge (Xetu)

๐Ÿ–ฅ๏ธ Interactive Web Interface (app.py)

Xuwor includes a built-in, zero-dependency browser interface for instant deployment:

python app.py

Open http://127.0.0.1:7860 in any web browser to access live voice recording, LLM thinking, speech generation, and bidirectional translation.


๐Ÿ›๏ธ Ecosystem Specifications

  • Compound System Architecture: Full-Duplex Conformer-ASR + Dense-LLM + Flow-Matching-TTS + Seq2Seq
  • Total Combined Parameters: ~3.15 Billion Parameters
  • Peak Operational Memory: ~2.30 GB - 3.80 GB RAM (CPU Edge-Calibrated)
  • Primary Languages: Assamese (as / เฆ…เฆธเฆฎเง€เฆฏเฆผเฆพ) & English (en)
  • Deployment Targets: Standalone Edge PC, Private Cloud Server, Enterprise Voice Gateway
  • Primary Maintainer: Axom AI Engineering Team

โš–๏ธ Proprietary License & Confidentiality

Copyright ยฉ 2026 Axom AI. All rights reserved.
This system, its orchestrated foundation weights, voice profiles, translation dictionaries, and architectural pipeline are the exclusive proprietary intellectual property of Axom AI. Unauthorized cloning, scraping, reproduction, or redistribution is strictly prohibited.

Downloads last month
-
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support