FAQs

Transcribing an hour of audio typically takes less than an hour. However, the duration may vary based on audio quality; softer or mumbled audio may require a bit more time.

Yes, each live transcription session is limited to two hours.

No, Transcribe does not set a limit on the number of transcriptions a user can submit. However, users are expected to use the service appropriately and fairly. If you have a large number of transcription tasks to process in bulk, please contact us separately at transcribe@tech.gov.sg.

You may refer to the tips for online meeting here for more information on the setup.

The performance and accuracy of speech-to-text transcription are typically evaluated using the Word Error Rate (WER), which summarizes errors in a transcript, including insertions, deletions, and substitutions. In practice, accuracy can vary based on the specific use case and is influenced by several factors, such as fluency, enunciation, and speakers’ accents. In formal speech settings with professional audio systems and articulate speakers, the WER can be as low as 10%. However, for meetings and focus group discussions featuring multiple speakers with varying accents and fluency, along with poor audio systems and significant background noise, the WER may rise to 30% to 40%. Consequently, accuracy can differ across transcription tasks, leading to some transcripts having more errors than others.

Transcribe does not support automatic language detection or switching within a single task. Once a language is selected for the task, the transcript will only be in that chosen language. Any other languages used during the task will not be transcribed.