Set gRPC Channel Config for google-cloud-speech Operation (4MB max_receive_message_length limit)

[brianarpie]

I want this functionality for google-cloud-speech:
42ee401

I keep hitting the 4MB limit when enabling speaker diarization for long audio files (because every speech recognition result includes word info from the beginning of the audio).

I got to this point from here:
https://issuetracker.google.com/issues/113502127

I can't find the creation of the grpc channel from google-cloud-speech. If it's not possible, my last resort is to fork grpc and change the defaults from there.

Any help much appreciated, thanks!

[blowmage]

Hi @brianarpie, I will show you how to construct a gRPC channel object, so you can specify channel arguments, but I do not think this will help. Speech has some specific content limits and changing the connection configuration won't make the API behave differently.

Here is how to construct a channel object with arguments:

require "google/cloud/speech/v1"

credentials = Google::Cloud::Speech::V1::Credentials.default
host = Google::Cloud::Speech::V1::SpeechClient::SERVICE_ADDRESS
chan_args = { "grpc.max_send_message_length"    => -1,
              "grpc.max_receive_message_length" => -1 }
chan_creds = GRPC::Core::ChannelCredentials.new.compose \
             GRPC::Core::CallCredentials.new credentials.client.updater_proc
channel = GRPC::Core::Channel.new host, chan_args, chan_creds

speech_client = Google::Cloud::Speech::V1.new credentials: channel

[brianarpie]

Thank you for the swift response @blowmage, this is exactly what I needed !

I'm well aware of the content limits. I'm hitting the 4MB message limit with audio files well below the 180 minute audio length limit. I can transcribe long audio files no problem, except when I enable speaker diarization. I'm fairly certain it's because of the extremely large payload response, due to the way word_info on the speech recognition result / speech recognition alternative is handled as I mentioned.

[brianarpie]

@blowmage thanks again, your snippet fixed the issue for us.
Watch out for this message limit for long audio files w/ speaker diarization enabled... I'm probably not going to be the last one to hit this.