Google-cloud-node: [cloud speech] the right way to get a transcript from operation? base64 decode!?

[mfkenson]

Though I couldn't find the usage of operation reference to get a transcript result.
I found something similar by base64-decoding operation.metadata.response.value..

The strange thing is I got some special characters after decoding the value.
Could you share the the right way to get a transcript from operation?

var operation = speechClient.operation('68850831366825');
operation.get({},function(err,res){
              if (err){
                  console.log(err);
                  return;
              }
              var metadata = res.metadata;
              var isDone = metadata.done;
              if (isDone && metadata.response) {
                  var b64string = metadata.response.value;
                  var buf = new Buffer(b64string, 'base64');
                  var transcript = buf.toString('utf8');
                  .....
             }

[stephenplusplus]

Just confirming the use case-- you are trying to get the transcript from a recognition request that was run in another process at another time?

[mfkenson]

@stephenplusplus Yes. thats what I am trying to do.

[stephenplusplus]

Thanks. We haven't designed an ideal solution for this, so the way you've been doing it would be the most correct. Off hand, I'm not sure what a good API would look like for this, so feel free to make suggestions / send a PR!

[stephenplusplus]

@lukesneeringer since Speech is about to go through the same process as Vision Partial Veneer, is this feature request relevant?

[ximus]

as a followup to to @mfkenson's snippet in the issue summary:

the base64 string found at metadata.response.value isn't usable data. It is the protocol buffers representation of the results. To get the result data, I am doing the following contortion to get useful data:

...
const results = speechClient.api.Speech.longrunningDescriptors.longRunningRecognize.responseDecoder(metadata.response.value).results
const result = map(results, r => r.alternatives[0].transcript).join('')

there may be a pretty way of doing this, but i've invested enough time in this for now.
This remains brittle solution as it relies on a deeply nested private API.

I find this public speechClient.operation rather useless in its current form and its presence on the public API more misleading than helpful as I couldn't find a public way of doing anything useful with the result set. Did I miss something? I certainly couldn't find any doc or examples showing me better.

I recommend you consider either deprecating it, augmenting it or putting a prominent warning about its limited use in the doc

[ximus]

I'd like to note that the use case it caters to (startRecognition in one process, poll for results in another) is an important use-case in my opinion.
Longer audio files can take real long to process by the speech API. I'm using Cloud Functions to startRecognition and I can't afford to have a function sit around waiting for the promise to resolve. Also, I just run into the max runtime quota for those and the function instance is forcibly terminated.

[mfkenson]

@ximus Even I could afford to sit and wait, there will be timeout exception (grpc)

[eoogbe]

Hi @mfkenson,
Thank you for reporting this.

Since the time that you reported this issue, we released an entirely new version of google-cloud-speech with significant alterations. Could you upgrade to the newest version (note that code changes will be necessary to use the new thing; your existing code will use the former, now deprecated code path), and it should fix this problem for you.

I am going to close this, but feel free to re-open if the issue persists.

[eoogbe]

I am no longer on the team