3 ms·
Cool! Which STT library/service are you using? How are you finding the error rate?
by abrichr 10y ago
Cool! Which STT library/service are you using? How are you finding the error rate?
- Emc2fma 10y agoGlad you like it! Have you had a chance to try it out yet? Currently I'm using Google's Speech API. Based on my tests, they have the best transcription results by far. I experimented with a few others (IBM, for one) and found that while the coding was easier (because they allow a wider variety of audio file formats), the results weren't as accurate. Google, on the other hand, is fairly picky about what types of files can be processed - this is also probably what makes them more accurate. So in the background, there's a fair bit of audio transcoding that goes on before transcription. What do you mean by error rate?
- abrichr 10y agoI haven't tried it out yet. By error rate I mean Word Error Rate (WER): https://en.wikipedia.org/wiki/Word_error_rate https://en.wikipedia.org/wiki/Word_error_rate
- Emc2fma 10y agoI can't give you an exact number because the audio/transcripts for users are completely hidden from human eyes/ears. Based on my own personal experience however, I would say I've gotten around 85% accuracy.