Qwen3 Forced Align:
Qwen3ForcedAlign is a sophisticated node designed to perform text-to-speech alignment, providing precise timestamps for each segment of text in relation to the corresponding audio. This node is particularly beneficial for applications requiring accurate synchronization between spoken words and their textual representation, such as in subtitling, language learning tools, or audio transcription services. By leveraging advanced forced alignment techniques, Qwen3ForcedAlign ensures that each word or sentence in the text is matched with its exact timing in the audio, enhancing the clarity and usability of audio-visual content. The node is capable of handling multiple languages, with a default focus on Chinese, and can segment text by sentences to improve alignment accuracy. Its integration into the Qwen3-ASR framework allows for seamless operation on CUDA-enabled devices, ensuring efficient processing even for large audio files.
Qwen3 Forced Align Input Parameters:
aligner
The aligner parameter specifies the Qwen3 ForcedAligner model to be used for the alignment process. This model is responsible for analyzing the audio and text inputs to determine the precise timing of each segment. The aligner must be pre-loaded and compatible with the Qwen3-ASR framework to ensure accurate results.
audio
The audio parameter represents the audio input that will be aligned with the provided text. It should be in a format that includes both the waveform and the sample rate. The waveform is typically a numerical representation of the audio signal, and the sample rate indicates how many samples per second are used to represent the audio. This parameter is crucial as it directly influences the alignment accuracy.
text
The text parameter is the string input that contains the textual content to be aligned with the audio. It supports multiline text, allowing for comprehensive alignment of longer passages. The text should be clear and well-structured to facilitate accurate segmentation and alignment.
language
The language parameter specifies the language of the text and audio inputs. It supports multiple languages, with a default setting of Chinese. This parameter ensures that the alignment process takes into account language-specific characteristics, which can significantly impact the accuracy of the alignment.
segment_by_sentence
The segment_by_sentence parameter is a boolean option that determines whether the text should be segmented by sentences during the alignment process. When set to true, the node will attempt to align each sentence individually, which can improve the precision of the alignment, especially in complex or lengthy texts. The default value is true.
Qwen3 Forced Align Output Parameters:
timestamps
The timestamps output provides a string containing the start and end times for each segment of text in relation to the audio. This output is essential for applications that require precise timing information, such as subtitle generation or detailed transcription.
text_list
The text_list output is a string that lists all the text segments that have been aligned. This output allows users to verify which parts of the text were successfully aligned and can be used for further processing or analysis.
start_times
The start_times output provides a string of the start times for each aligned text segment. This information is crucial for understanding when each segment begins in the audio, enabling precise synchronization.
end_times
The end_times output provides a string of the end times for each aligned text segment. This output complements the start times, offering a complete picture of the duration and timing of each text segment within the audio.
Qwen3 Forced Align Usage Tips:
- Ensure that the audio input is clear and of high quality to improve alignment accuracy.
- Use the
segment_by_sentenceoption for longer texts to enhance precision by aligning each sentence individually. - Verify that the language parameter is set correctly to match the language of the text and audio inputs for optimal results.
Qwen3 Forced Align Common Errors and Solutions:
Model not found, downloading: <model_name>
- Explanation: This error occurs when the specified model is not found locally and needs to be downloaded.
- Solution: Ensure that you have a stable internet connection to allow the model to be downloaded successfully. Check the model name for any typos.
Using local model: <local_model_path>
- Explanation: This message indicates that the node is using a locally stored model for alignment.
- Solution: No action is needed if the local model is correct. If issues arise, verify the model's integrity and compatibility.
Alignment completed, <number> segments
- Explanation: This message confirms that the alignment process has finished and indicates the number of segments aligned.
- Solution: Review the output to ensure that all expected segments are included. If discrepancies are found, check the input text and audio for errors.
