Machines will take over human – in transcription

We will soon witness a day when the machines are way better transcribers than we humans. The improvement in machine accuracy during last few years has been huge.

Google, as one of the industry leaders, has reported almost 20 % improvement in accuracy of its speech recognition technology during the last four years. In May 2017 Google announced that their Speech API has reached accuracy of 95,1 %, which is the first time ever for a machine to surpass the threshold of human accuracy of 95,0 %.

Steady growth in speech recognition accuracy

As you can see from the graph below, the improvement in accuracy during the past years has been extremely fast. Although now the growth has become relative stable, the accuracy is still improving slowly and steadily. There is always room for improvement, although figures over 95 % are super good already.

The figures shown here are measured with American English accent which currently has the highest accuracy. Other languages do not yet reach the same accuracy, although the improvement is also strong with other languages and accents.

Variety of use-cases becomes wider

The variety of speech recognition use-cases becomes wider as the speech recognition accuracy reaches the human threshold.

We will tell more about the different use-cases in our upcoming blog posts, so stay tuned!

Decrease in Word Error Rate

As the Word Error Rate has significantly decreased during the past years (check the graph below), different industry players have started to utilize speech recognition technology in many ways.

Spoken ONLINE utilizes automated speech recognition technology for transcribing especially long audio files such interview recordings. There are several other players in the field using different technologies various transcription purposes. Only time will show which technologies are the best in the long run, but at the moment Google Speech API is the superior based also on academic research.

We at Spoken ONLINE are confident that Google is the number one player in speech recognition technology also in the future. Therefore, we are extremely happy to offer our service with Google Speech API providing the core technology for our transcription tool.