Final transcripts showing empty transcription

AI & Data Science Deep Learning (Training & Inference) Riva
1

Please provide the following information when requesting support.

Hardware - GPU (T4)
Hardware - CPU
Operating System
Riva Version: 2.6.0
TLT Version (if relevant)
How to reproduce the issue ? (This is for errors. Please share the command and the detailed log here)

Hello everyone!
I’ve created my own NeMo model and then all other steps of riva model, rmir model and riva repositories according to documentation.
I’m using Quick Start Scripts to Deploy model. After using this line: riva_streaming_asr_client --audio_file <wav file location>, I’m not getting any transcription.
Here is the output I’m getting:

I1031 08:47:32.209681   109 riva_streaming_asr_client.cc:154] Using Insecure Server Credentials
Loading eval dataset...
filename: <wav file location>
Done loading 1 files
File: <wav file location>

Final transcripts:

Audio processed: 7.34695e-40 sec.

Not printing latency statistics because the client is run without the --simulate_realtime option and/or the number of requests sent is not equal to number of requests received. To get latency statistics, run with --simulate_realtime and set the --chunk_duration_ms to be the same as the server chunk duration
Run time: 4.68129 sec.
Total audio processed: 286.728 sec.
Throughput: 61.2499 RTFX

So, as you can see Final transcripts response is empty.

From docker log, I’m getting this:

  > Triton server is ready...
I1031 04:19:51.898931   423 riva_server.cc:120] Using Insecure Server Credentials
I1031 04:19:51.944115   423 model_registry.cc:110] Successfully registered: citrinet-1024-en-US-asr-streaming for ASR
W1031 04:19:51.961644   423 grpc_riva_asr.cc:157] citrinet-1024-en-US-asr-streaming has no configured wfst normalizer model 
I1031 04:19:51.980005   423 riva_server.cc:160] Riva Conversational AI Server listening on 0.0.0.0:50051
W1031 04:19:51.980062   423 stats_reporter.cc:41] No API key provided. Stats reporting disabled.
I1031 08:47:00.767860   428 grpc_riva_asr.cc:892] ASRService.StreamingRecognize called.
I1031 08:47:00.768599   428 grpc_riva_asr.cc:919] ASRService.StreamingRecognize performing streaming recognition with sequence id: 1779700260
I1031 08:47:00.800891   428 grpc_riva_asr.cc:976] Using model citrinet-1024-en-US-asr-streaming for inference
I1031 08:47:00.801008   428 grpc_riva_asr.cc:992] Model sample rate= 16000 for inference
I1031 08:47:00.848378   428 riva_asr_stream.cc:214] Detected format: encoding = 1 numchannels = 1 samplerate = 16000 bitspersample = 16
I1031 08:47:03.263229   428 grpc_riva_asr.cc:1093] ASRService.StreamingRecognize returning OK
I1031 08:47:32.256738   428 grpc_riva_asr.cc:892] ASRService.StreamingRecognize called.
I1031 08:47:32.257011   428 grpc_riva_asr.cc:919] ASRService.StreamingRecognize performing streaming recognition with sequence id: 2124845530
I1031 08:47:32.257077   428 grpc_riva_asr.cc:976] Using model citrinet-1024-en-US-asr-streaming for inference
I1031 08:47:32.257154   428 grpc_riva_asr.cc:992] Model sample rate= 16000 for inference
I1031 08:47:32.257484   428 riva_asr_stream.cc:214] Detected format: encoding = 1 numchannels = 1 samplerate = 16000 bitspersample = 16
I1031 08:47:36.936904   428 grpc_riva_asr.cc:1093] ASRService.StreamingRecognize returning OK

Can somebody direct me on how to comprehend what went wrong?

2

Hi @shuvohishab

Thanks for your interest in Riva,

Request to share the following details,

  1. Base Nemo model used (with NGC link)
  2. If any commands used for retraining the custom model (if any, if directly downloaded and converting to riva this won’t be there)
  3. Riva build command used (complete)
  4. Riva Deploy command (complete)

Thanks

3

Also experiencing this with pretrained citrinet and flashlight decoder

4

Hi @francois20

Can you kindly share the following details with us

  1. pretrained citrinet ngc model link
  2. Riva build command used (complete)
  3. Riva Deploy command (complete)
5

Hi guys, thanks for your guidance and responses.

I’ve solved the issue by providing --language_code=<provided language code in riva-build stage>.

6

Hey @francois20 , don’t forget to provide --name and --language_code almost identical to each other, in riva-build stage. The working pattern of --name argument in my case was: conformer-<language_code>-asr-streaming.
Don’t know about the behind the scene factors, but it worked for me.

And, at the time of inference, don’t forget to provide --language_code=<your provided language code in riva-build stage>.

Thanks.

7

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.