ComfyUI Node: SenseVoice 语音识别

SenseVoice 语音识别:

SenseVoiceNode is a powerful tool designed for speech recognition, enabling the conversion of audio input into text. This node is part of the ComfyUI-FunAudioLLM suite and leverages advanced models to process and transcribe spoken language efficiently. It is particularly beneficial for applications requiring real-time or near-real-time speech-to-text conversion, such as voice-controlled interfaces, transcription services, and accessibility tools. The node supports various configurations to optimize performance based on the length of the audio and the need for punctuation segmentation. By utilizing sophisticated models and techniques, SenseVoiceNode ensures high accuracy and reliability in recognizing and transcribing speech, making it an essential component for developers and AI artists looking to integrate voice recognition capabilities into their projects.

SenseVoice 语音识别 Input Parameters:

audio

The audio parameter is a dictionary containing the audio data to be processed. It includes the waveform of the audio and its sample rate. This parameter is crucial as it provides the raw input that the node will transcribe into text. The waveform should be properly formatted and sampled to ensure accurate transcription results.

use_fast_mode

The use_fast_mode parameter is a boolean that determines whether the node should operate in a faster processing mode. When set to True, the node processes audio more quickly but may impose restrictions on the length of the audio input, specifically if it exceeds 30 seconds. This mode is ideal for applications where speed is prioritized over handling longer audio segments.

punc_segment

The punc_segment parameter is a boolean that indicates whether punctuation segmentation should be applied to the transcribed text. When enabled, the node uses a punctuation model to enhance the readability of the output by adding appropriate punctuation marks. This is particularly useful for generating more natural and human-readable transcriptions.

SenseVoice 语音识别 Output Parameters:

rich_transcription

The rich_transcription output parameter provides the transcribed text from the audio input, enhanced with punctuation and other post-processing features. This output is the final result of the node's processing and is designed to be as accurate and readable as possible, making it suitable for direct use in applications that require text representation of spoken language.

SenseVoice 语音识别 Usage Tips:

• To optimize performance for short audio clips, enable use_fast_mode to reduce processing time while ensuring the audio length does not exceed 30 seconds.
• Enable punc_segment to improve the readability of the transcribed text by adding punctuation, which is especially useful for creating transcripts that are easy to understand and follow.

SenseVoice 语音识别 Common Errors and Solutions:

Audio length is too long, please set use_fast_mode to False.

• Explanation: This error occurs when the audio input exceeds the length limit of 30 seconds while use_fast_mode is enabled.
• Solution: Disable use_fast_mode to allow processing of longer audio segments, or shorten the audio input to meet the length requirement.

Model loading error

• Explanation: This error might occur if there is an issue with downloading or loading the required model files.
• Solution: Ensure that the model files are correctly downloaded and accessible. Check your internet connection and the file paths specified in the configuration.