Audio

notion image
In the input fields "Title (en)", "Title (ru)", and "Title (ua)", fill in the titles according to the respective languages.
💡
Fill in Title (ru) and Title (ua) only if you want to get a human readable localized class name.
To read more about attribute interpolation and available options go to the Attribute interpolation section.
For ID export, enter the name you would like to use while exporting.
There are also two checkboxes:
  1. Read only - if selected, the attribute cannot be used to select answer options.
  2. Private - if selected, this will make the attribute private without exporting it in json.
💡
Unlike audio added to comments, audio attributes are part of the annotation data and therefore support versioning, Undo/Redo, and export.
After recording:
  1. The audio is saved as the attribute value.
  2. The recording is automatically transcribed.
  3. The generated transcript is displayed in the attribute.
  4. The transcript can be edited if necessary.

Speech Recognition Engines

Voice recordings can be transcribed using one of the available speech recognition engines:
  • Whisper;
  • Gemini;
  • Chirp.
The available engines depend on the platform configuration.

Maximum Audio File Size

Administrators can limit the maximum size of audio files used in a project.
Configure the maximum allowed audio size in the Project settings.
This setting helps control storage

Recording a Voice Note

Annotators can record a voice note directly in the editor using the configured Audio attribute.
Voice recording can also be performed while the video is playing at a reduced playback speed:
  • 0.5×
  • 0.75×
This is useful for dynamic scenes where it may be difficult to describe everything at the video's normal playback speed.
After the recording is completed, the voice annotation is automatically synchronized with the video's original playback speed.

Audio Attribute Playback

When an Audio attribute contains multiple voice clips, the editor provides the Play all parts with video button.
Clicking the button once starts automatic playback of all voice clips belonging to the attribute.
The clips are:
  • played in frame order;
  • synchronized with the video;
  • played without manually switching between individual clips.
While the voice clips are being played, the video's original audio is temporarily muted.
Pauses between voice clips are skipped.
This allows the complete voice annotation of an object to be listened to continuously, without manually opening and playing each individual clip.

Voice Note Duration

A single voice note can be up to 45 minutes long.
The increased duration allows an annotator to record the narration for an entire video in a single pass.

Positioning of Voice Notes

A voice note is anchored to the frame where the recording was started.
The note remains associated with that starting frame rather than being moved to the position where the video was when the recording was stopped.
This ensures that the voice note remains associated with the point in the video where the recording began.