Shanmugapriyan/Qwen3-Omni-30B-A3B-Instruct-merged-ft

Qwen3 Omni 30B ASR Fine-Tuned Model

Model Description

This model is a fine-tuned version of Qwen3-Omni-30B-A3B-Instruct designed for Automatic Speech Recognition (ASR) tasks.

The model was fine-tuned on multilingual speech data containing English and Hindi audio transcripts to improve transcription accuracy.

It can transcribe spoken audio into text and is suitable for:

• Voice assistants
• Call transcription
• Conversational AI systems
• Speech analytics pipelines

---

Base Model

Base model used:

Qwen/Qwen3-Omni-30B-A3B-Instruct

This model supports multimodal inputs including:

• Text
• Audio

---

Fine-Tuning Details

Property	Value
Base Model	Qwen3 Omni 30B
Task	Automatic Speech Recognition
Languages	English, Hindi
Training Method	Fine-tuning
Format	Instruction-based training
Dataset Format	JSONL

---

Dataset

The model was trained on a dataset containing:

• Audio recordings
• Corresponding transcripts
• Language labels
• Duration metadata

Languages included:

• English
• Hindi

Dataset repository:

Shanmugapriyan/qwen_omni_ft_data

---

Usage

You can use this model for Automatic Speech Recognition.

---

Intended Use

This model is designed for:

• Speech-to-text transcription
• Conversational AI pipelines
• Voice assistant systems
• Customer call analytics

---

Limitations

• Requires high GPU memory (30B model)
• Accuracy depends on audio quality
• Performance may vary on unseen languages

---

Hardware Requirements

Recommended hardware:

• GPU: A100 / H100
• VRAM: 40GB+

---

Author

Shanmugapriyan T

AI Engineer | Voice AI Systems | LLM Fine-tuning

---

License

Apache 2.0