Async call does not return the entire transcript

[davidmeza1]

Taking a wild shot here hoping you have seen this. I am trying to transcribe a 2+ minute audio file asynchronously. I am able to send the audio file and return the results. However, I am only getting 18 seconds back. Checking the google API dashboard, I see the entire 2 minute file is processed. I have checked with google support and all looks well there.

Have you seen any issues with not all of the file being returned. I am running R 3.4.1 and googleLanguageR 0.1.0.9000 on a MAC 10.12.6

Testwav3 <- ("gs://qual_audios/audio_only3.flac")
Testasync3 <- gl_speech(Testwav3, encoding = "FLAC", sampleRateHertz = 16000L, asynch = TRUE)
Test_result3 <- gl_speech_op(Testasync3)

[MarkEdmondson1234]

Hmm, well it could be the API in which case I can't do much aside recommend raising a bug report with them, but I would also check the encoding of the wave file, as I sometimes got back a shorter file due to it thinking 8-bit was 16-bit, for example. However if you are getting some audio back and that plays ok, its not that.

If you can, perhaps check by splitting up the audio file into smaller chunks that the non-asnch call can deal with (less than 60 seconds) - something like Audacity should deal with that.

[jenswaeckerle]

We are having a similar issue here: We tried audio from videos that youtube is able to provide subtitles for, so we think the speech recognition is probably not the problem. How do I check which encoding is correct, should I save the audio file as 8-bit or 16-bit as the function itself can't specify this? Do zou have any other ideas how we could circumvent this problem?

I'm running R 3.4.1 on Mac 10.12.6, using a FLAC file.

Thanks!

[MarkEdmondson1234]

I need to test whether this is an issue with this library or the API. The encoding and sample rate should match the audio file, although in some cases it autodetects.

[MarkEdmondson1234]

Ok it was the library, I didn't test it with a long enough sound file. It is fixed now, and I changed the output structure a little to make it easier to get the two versions of output.

The return is now a list of two data.frames (tibbles) - one as $transcript holds a transcript and any number of alternatives you specify with the confidence in the transcription; the other $timings carries information on when the words were spoken in the audio.

[MarkEdmondson1234]

@jenswaeckerle @davidmeza1 This fix is now on CRAN v 0.1.1 - thanks for the report!