How to Improve Transcription Accuracy
Everything you need to know about Ebby's transcription accuracy and how to improve it
Ebby's transcription accuracy depends on the quality of the audio being transcribed.
Our speech to text algorithm is not flawless, but will reach 99% accuracy on clear, well-recorded audio - not that bad, considering it costs x10 less and delivered in minutes ;-)
💡 Tip: Transcription accuracy varies for each recording
You can preview a recording before transcribing it to get a glimpse of the transcription accuracy or compare different dialects.
A preview will transcribe the first 2 minutes of your recording.
If you're in need of extra previews, please contact us
What does Ebby struggle with?
A basic rule of thumb is: if you have a hard time understanding what's being said, so would Ebby.
Nevertheless, Ebby can perform magic while applying different mechanisms to reduce noise and handle low volume, for example.
Background noise, music and echoes
Although our algorithm applies noise reduction techniques to your file, excessive background noise during recording may negatively affect the accuracy of the transcript.
Cross-talk may be interpreted as background noise and ignored.
Background music is also considered as "noise" and speech may be ignored - songs/lyrics will not be accurately transcribed.
- Try and record in a noise-free environment as much as possible. Speech should sound clear, without distortion or unexpected noise.
- Use good quality and well-positioned microphone, particularly when background noise is present.
- Try to avoid multiple people talking at the same time, or at different volumes.
- Avoid background music - upload your original recording without background music or sound effects (mainly for video and podcast transcription).
- Applying noise-reduction signal processing to the audio typically reduces recognition accuracy (see Technical aspects), but in extreme cases, consider cleaning up your recording using an audio/video editor (Audacity is free and has noise reduction capabilities).
Volume
Low volume recording may reduce accuracy, especially if a lossy codec is used.
Multiple people talking at different volumes, or at the same time may be interpreted as background noise and ignored.
- It's best to provide audio that is as clean as possible by using a good quality and well-positioned microphone.
Dialect and Accent
Strong accent or regional dialect may decrease transcription accuracy.
- Ebby supports various dialects for many languages - make sure to choose the right dialect when transcribing.
- For multiple speakers with different dialects, choose the most prominent in your recording.
Speech Rate
How fast do people talk in your recording?
- Adopt a conversational speech rate or even slow down if possible.
Technical aspects
Technical recording recommendations
-
Upload your original recording: avoid re-sampling, clipping, converting or transcoding your file as it reduces transcription accuracy.
We handle a wide range of file types and encodings. If your file format is not supported let us know and we'll add it in no time. - Use a lossless codec to record, FLAC or LINEAR16 is recommended.
- Optimal sampling rate is 16,000 Hz, higher sampling rate is not an issue but low sampling rate may reduce accuracy.
- Do not use automatic gain control (AGC).
- All noise reduction processing should be disabled.
Try for free