ComfyUI Node: IndexTTS2 Emotion From Text

IndexTTS2 Emotion From Text:

The IndexTTS2EmotionFromText node is designed to analyze a given text input and generate an emotion vector that represents the emotional content of the text. This node leverages the QwenEmotion model to infer emotions from text, providing a structured way to quantify emotions such as happiness, anger, sadness, fear, disgust, melancholy, surprise, and calmness. The primary goal of this node is to facilitate the integration of emotional context into AI-driven applications, particularly in the realm of text-to-speech synthesis, where conveying the right emotional tone is crucial. By converting textual descriptions into a numerical emotion vector, this node enables more nuanced and expressive audio outputs, enhancing the overall user experience.

IndexTTS2 Emotion From Text Input Parameters:

text

The text parameter is a string input that represents the text from which emotions are to be inferred. It is crucial for this text to be descriptive enough to allow the QwenEmotion model to accurately detect and quantify the emotions present. The text should not be empty or consist solely of whitespace, as this will result in an error. There are no explicit minimum or maximum length constraints provided, but the text should be sufficiently detailed to capture the intended emotional nuances.

IndexTTS2 Emotion From Text Output Parameters:

emotion_vector

The emotion_vector is a list of floating-point numbers that represents the detected emotions in a structured format. Each number corresponds to the intensity of a specific emotion, following a predefined order: happy, angry, sad, afraid, disgusted, melancholic, surprised, and calm. The values are clamped between 0.0 and 1.4 to ensure consistency and prevent any single emotion from dominating the vector. This output is essential for applications that require a quantitative representation of emotions, allowing for precise control over emotional expression in AI-generated content.

info

The info output is a string that provides a summary of the detected emotion vector, including the sum of the vector values and the individual intensities of each emotion. This information is useful for understanding the overall emotional profile of the text and for debugging purposes, as it offers insight into how the text was interpreted by the model.

IndexTTS2 Emotion From Text Usage Tips:

• Ensure that the input text is descriptive and context-rich to allow for accurate emotion detection. Avoid using vague or ambiguous language.

• Use the info output to verify the detected emotions and adjust the input text if necessary to achieve the desired emotional profile.

• Be mindful of the emotion vector's sum, which should not exceed 1.5. If it does, consider reducing the intensity of certain emotions in the text or using the Emotion Vector node to adjust the values.

IndexTTS2 Emotion From Text Common Errors and Solutions:

Text is empty. Please provide descriptive emotion text.

• Explanation: This error occurs when the input text is either empty or consists only of whitespace.
• Solution: Provide a non-empty, descriptive text input that clearly conveys the emotions you wish to detect.

QwenEmotion model not found at: <path>

• Explanation: The specified directory for the QwenEmotion model does not exist, indicating that the model files are missing or the path is incorrect.
• Solution: Verify that the QwenEmotion model is correctly installed and that the path specified in the node configuration is accurate.

Emotion vector sum <value> exceeds maximum 1.5. Reduce intensities or adjust with the Emotion Vector node.

• Explanation: The sum of the emotion vector values exceeds the allowed maximum of 1.5, which can lead to an imbalance in the detected emotions.
• Solution: Adjust the input text to reduce the intensity of certain emotions or use the Emotion Vector node to manually adjust the vector values to ensure the sum does not exceed 1.5.