AuK Whisper Transcribe:
The AuKWhisperTranscribe node is designed to transcribe audio into text, which can be used to craft edit instructions or verify the output of AuK processes. This node leverages local models stored under models/whisper or models/audio_encoders, ensuring that transcription tasks can be performed offline. However, it also offers the flexibility to download a known Whisper checkpoint from Hugging Face if explicitly enabled, providing access to a broader range of models. The primary goal of this node is to facilitate seamless audio-to-text conversion, which is crucial for tasks that require precise audio editing or transcription. By converting spoken language into text, it allows users to generate accurate instructions for audio modifications or to ensure that the audio output aligns with the intended content.
AuK Whisper Transcribe Input Parameters:
audio
This parameter accepts the audio file that you wish to transcribe. It serves as the primary input for the transcription process, allowing the node to convert the spoken content into text. The quality and clarity of the audio can significantly impact the accuracy of the transcription.
model
This parameter specifies the Whisper model to be used for transcription. You can choose from models available in the models/whisper or models/audio_encoders directories. If the selected model is not available locally, it can be downloaded if the download_if_missing option is enabled. This flexibility allows you to select a model that best suits your transcription needs.
download_if_missing
This boolean parameter determines whether the node should download the selected model from Hugging Face if it is not available locally. By default, this is set to False, keeping the node fully offline. Enabling this option allows you to access a wider range of models without manually downloading them.
model_size
This parameter defines the size of the model to be used when download_if_missing is enabled and the selected model is not available locally. Options include sizes like tiny, base, small, medium, large-v3, and large-v3-turbo. The default is large-v3, which provides a balance between performance and resource usage.
language
This parameter provides a hint about the spoken language in the audio. It supports multiple languages, including auto for automatic detection, english, chinese, german, spanish, russian, korean, french, japanese, portuguese, turkish, polish, and catalan. Correctly setting this parameter can improve transcription accuracy.
task
This parameter specifies the task to be performed: transcribe or translate. The transcribe option keeps the spoken language as is, while translate converts the audio content into English. This allows you to choose between maintaining the original language or translating it for broader accessibility.
AuK Whisper Transcribe Output Parameters:
text
The output of the AuKWhisperTranscribe node is a string containing the transcribed text from the input audio. This text can be used for further processing, such as generating edit instructions or verifying the accuracy of audio outputs. The quality of the transcription depends on the clarity of the audio and the appropriateness of the selected model and language settings.
AuK Whisper Transcribe Usage Tips:
- Ensure that the audio input is clear and free from excessive background noise to improve transcription accuracy.
- Select the appropriate model and language settings to match the characteristics of your audio input for optimal results.
- If you need to transcribe audio in a language not supported locally, consider enabling the
download_if_missingoption to access additional models from Hugging Face.
AuK Whisper Transcribe Common Errors and Solutions:
"Whisper requires non-empty audio."
- Explanation: This error occurs when the input audio file is empty or not properly loaded.
- Solution: Verify that the audio file is correctly loaded and contains audible content before attempting transcription.
"Model not found on disk and download_if_missing is disabled."
- Explanation: This error indicates that the selected model is not available locally, and the option to download missing models is turned off.
- Solution: Enable the
download_if_missingoption to allow the node to download the required model from Hugging Face, or manually download and place the model in the appropriate directory.