Qwen2-Audio

import numpy as np
from versa import qwen2_model_setup, qwen2_speech_emotion_metric

# Setup model (do this once)
qwen_utils = qwen2_model_setup()

# Load your audio
audio = np.random.random(16000)  # Replace with actual audio

# Analyze emotion
result = qwen2_speech_emotion_metric(
    qwen_utils=qwen_utils,
    pred_x=audio,
    fs=16000
)

print(f"Emotion: {result['qwen_speech_emotion']}")

import numpy as np
from versa import (
    qwen2_model_setup,
    qwen2_speaker_count_metric,
    qwen2_speaker_gender_metric,
    qwen2_speech_emotion_metric,
    qwen2_recording_quality_metric
)

# Setup once
qwen_utils = qwen2_model_setup()
audio = np.random.random(48000)  # Replace with actual audio
fs = 48000

# Run multiple analyses
analyses = {
    "speaker_count": qwen2_speaker_count_metric(qwen_utils, audio, fs),
    "speaker_gender": qwen2_speaker_gender_metric(qwen_utils, audio, fs),
    "emotion": qwen2_speech_emotion_metric(qwen_utils, audio, fs),
    "quality": qwen2_recording_quality_metric(qwen_utils, audio, fs)
}

for metric, result in analyses.items():
    print(f"{metric}: {result}")

import numpy as np
from versa import qwen2_model_setup, qwen2_base_metric

qwen_utils = qwen2_model_setup()
audio = np.random.random(16000)

# Use custom prompt
custom_prompt = """Analyze this audio and describe the speaker's 
emotional state in detail, including any subtle nuances."""

result = qwen2_base_metric(
    qwen_utils=qwen_utils,
    pred_x=audio,
    fs=16000,
    custom_prompt=custom_prompt,
    max_length=1000
)

print(result)

pip install transformers accelerate