Qwen2-Audio

Qwen2-Audio is a multimodal audio-language model that handles more than transcription. In MLX Audio it can be used for ASR, translation, captioning, emotion recognition, and general audio understanding through text prompts.

Models

Model	Quantization	HuggingFace
Qwen/Qwen2-Audio-7B-Instruct	bf16	Model Card
mlx-community/Qwen2-Audio-7B-Instruct-4bit	4-bit	Model Card

Usage

CLIPython

python -m mlx_audio.stt.generate \
    --model mlx-community/Qwen2-Audio-7B-Instruct-4bit \
    --audio audio.wav \
    --prompt "Transcribe the audio."

from mlx_audio.stt.utils import load_model

model = load_model("mlx-community/Qwen2-Audio-7B-Instruct-4bit")

result = model.generate("audio.wav", prompt="Transcribe the audio.")
print(result.text)

Typical Prompts

• Transcribe the audio.
• Translate the speech to French.
• What emotion is the speaker expressing?
• Describe the environmental sounds in this clip.

Capabilities

• Speech transcription
• Speech translation
• Audio captioning
• Emotion and sentiment analysis
• Environmental sound classification
• General audio understanding

Links

• Source code
• In-repo README
• Qwen/Qwen2-Audio-7B-Instruct