Our machine-learning robots are pretty neat, Trinting your audio or video files into text - at lightening speeds. But they’re not miracle workers (just magicians).
We do expect editing on the other end, of course. Speech-to-text technology is by no means perfect. And the amount of editing you will be required to do is directly proportional to the quality of the audio, with reasonably quality audio yielding up to 95-98% accuracy.
The idea behind Trint is not to replace manual transcription, but to greatly speed up that process by doing a large share of the work for you. When we see people confused or disappointed by the accuracy of their transcription, it’s usually because of one of two things: 1) recording quality; or 2) heavy accents.
We can only be as good as the files you give us. Background noise, music, street sounds, hiss, echos, or distant microphones will all impact the quality of the transcript.
Here are a few tips to get help you get better audio and get the most out of Trint:
TRINT TIP 😎 VOCAB BUILDER: Using Vocab Builder can help increase the accuracy rates of your transcriptions! This feature allows you to create a custom dictionary for your Trints to use. You can add in proper names, industry specific terms, and a lot more! Anything that Trint typically misses can be added to the list. https://support.trint.com/hc/en-us/articles/360000961925-Vocab-Builder-FAQ
TRINT TIP 😎 AVOID BACKGROUND NOISE: Watch out for music, street noise or nearby conversations. If you can control the environment, choose a quiet place. In a noisy room, you want to keep the microphone a few inches from the speaker(s) if possible. With poor quality audio, you will often be able to hear a static/hissing noise. Try focusing on all of the sounds in your audio apart from human speech; is there a significant amount of extra sound? If so, you may want to try and improve the audio before running it through Trint.
TRINT TIP 😎 MAKE SURE THE MICROPHONE IS CLOSE TO THE SPEAKER(S): An iPhone recording a conversation between two people in a quiet room should be fine. But if there are eight people talking around a board table with one microphone in the center, you may be disappointed with the results.
TRINT TIP 😎 LISTEN FOR ECHO: A hollow sounding room will degrade the recording quality.
TRINT TIP 😎 ACCENTS & LANGUAGES: In addition, heavy foreign-accented English may be too difficult for the algorithms to decipher. Remember, Trint currently offers you three language models in English: North American, UK and Australian. We also have language models in European Spanish & French, German, Italian, Portuguese, Russian, Polish, Finnish, Hungarian, Dutch, Romanian and Swedish.
Trint can’t decipher every kind of audio, but with a little experimenting and a little extra effort, we are confident you will be really happy with the results.
TRINT TIP 😎 ONE MORE TIP: if you have speakers with multiple accents, we suggest you choose the transcription algorithm of 1) the weakest speaker; or 2) the key speaker (i.e. the interview subject rather than the interviewer.)
We're here to help
If you’re comfortable sharing your Trint with a member of our Support team, they’ll happily give you feedback. Send a note to us at firstname.lastname@example.org.