Speech Transcription with Whisper and Vosk

[minhquoc0712]

This branch integrates two speech transcription libraries: Whisper by openAI and Vosk.

Features:

• Adding target classes for the output of each transcription library.
• Integrate the download, load, infer and eval, etc, of the two libraries into OpenDR learner classes.
• Add a demo for the speech transcription task.
• Implement ROS node for the speech transcription task with two backbones, Whisper and Vosk.
• Implement ROS2 node for the speech transcription task.
• Add document for ros bridge, ros messages, transcription target, and learner class.

Notes:

• For ROS node, transcribing streaming speech is not native for Whisper. Therefore I add some heuristic rules for breaking the phrases, detecting silence, etc.
• For RO2, the audio common package seems to update the default audio topic to audio/audio. If that is the case, I think we need to change the default input audio topic for other audio-related ROS2 nodes.

Known issues:

• Currently, there is a problem with the ROS2 node. It cannot do the speech transcription properly. The reason could be the audio format published by the audio capture node in ROS2.

[minhquoc0712]

Hi, can you review my pull request? @omichel , @passalis ?

[omichel]

It seems there are some PEP8 errors, could you please address them?

/home/runner/work/opendr/opendr/projects/opendr_ws_2/src/opendr_perception/opendr_perception/speech_transcription_node.py:204:1: W293: blank line contains whitespace /home/runner/work/opendr/opendr/projects/opendr_ws/src/opendr_perception/scripts/speech_transcription_node.py:58:1: E303: too many blank lines (4)

[minhquoc0712]

Hi, I have fixed the styles